www.elsevier.com/locate/foreco

Forest Ecology and Management 224 (2006) 58–80

BARUMODEL: Combined Data Based Mechanistic models

of runoff response in a managed rainforest catchment

Nick A. Chappell a,*, Wlodek Tych a, Arun Chotai a, Kawi Bidin b,Waidi Sinun c, Thang Hooi Chiew d

a Lancaster Environment Centre, Lancaster University, Lancaster LA1 4YQ, UKb Universiti Malaysia Sabah, 88999 Kota Kinabalu, Sabah, Malaysia

c R&D Division, Yayasan Sabah, 88817 Kota Kinabalu, Sabah, Malaysiad Forestry Department Peninsular Malaysia, 50660 Kuala Lumpur, Malaysia

Abstract

The impact of different forestry practices onheadwater streams in thewet tropics is a serious environmental concern.Despite this, there arevery few

experimental catchment studies that quantify the hydrological impacts of specific tropical forestry operations.While new field studies are imperative,

more could be gained from existing catchment studies by extracting information from the available time-series. Data Based Mechanistic (DBM)

models can be used to extract hydrological information from such time-series by fitting a range of Transfer Function models using the simplified

recursive instrumental variable (SRIV) algorithm and without a priori assumptions about the water pathways. An optimal model and its associated

hydrological system parameters are then identified from objective statistical measures (e.g.R2t , YIC) and only then by selection of that model with the

most plausible physical explanation. While the DBM method has been applied to the relationship between rainfall input and streamflow output in

tropical rainforests, it has never been used to simultaneously examine the relationships within rainfall–streamflow data and the component water

pathwaysof overlandflow, subsurfaceflowand transpiration.Modelling these component pathways is important to theunderstandingof likely changes

in the rainfall–streamflowbehaviour resulting from tropical forestry.We aim to show thevalue of ourmultiple componentDBMmodels to simulate the

sensitivity of stream behaviour to different densities of skidder vehicle trails within a managed rainforest in Borneo.

The first application of DBMmodelling to the component water pathways of an equatorial rainforest catchment shows that the overall rainfall–

streamflow response is flashy, with a residence time of 36.25 � 0.19 min, when compared to other small catchments, but on minor aquifers. The

overall response is, however, much less flashy in comparison to the infiltration-excess overland flow pathway which has a residence time of only

5.50 � 0.09 min. The overland flow pathway is also less non-linear in its response in comparison to the subsurface pathway and overall response. A

new DBM model of transpiration was found, and was similarly shown to give a rapid dynamic response to the input variable, in this case air

temperature. The DBM approach indicated that higher order models (e.g. multiple subsurface pathways) were not statistically identifiable within

any of the high frequency, 12-month time-series modelled. The first application of DBM modelling to land use/change scenarios in the same

equatorial rainforest catchment suggested that the majority of the stream behaviour, in a period 7-years after selective logging, is insensitive to the

skid trail densities associated with reduced impact logging (RIL) or clearfell systems in Malaysia. This is largely because of the relative

insignificance of the overland flow pathway and the fact that changes in this pathway are predicted to occur on the rising stage of stream

hydrographs, rather than at the more critical peak. While controversial, this result is consistent with our previous findings which show that some

forestry impacts on hydrology are exaggerated in the popular perception.

# 2006 Elsevier B.V. All rights reserved.

Keywords: DBM model; Peakflow; Skid trail; time-series data; Transfer function; Tropical forestry

* Corresponding author. Tel.: +44 1524 593933; fax: +44 1524 593985.

E-mail address: n.chappell@lancaster.ac.uk (N.A. Chappell).

0378-1127/$ – see front matter # 2006 Elsevier B.V. All rights reserved.

doi:10.1016/j.foreco.2005.12.008

1. Introduction

1.1. Value of hydrological time-series data from managed

tropical rainforests

There are many popular perceptions about the hydrological

impacts of logging natural forests in thewet tropics (Bruijnzeel,

2004; Chappell, 2005), yet there are very few quantitative

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–80 59

time-series data on which to base management decisions

(Bruijnzeel, 1990, 1992, 2001; Chappell et al., 2004b; Thang

and Chappell, 2004). The spatial variability of the natural

water-pathways combined with the patchwork of impacts of

selective commercial forestry requires that significant change

to the hydrological system must be examined at an integrated

scale (Chappell et al., 2004b). To a hydrologist, this level of

integration is the 0.1–50 km2 scale of the ‘experimental

catchment’ (Gregory and Walling, 1973). Thus, changes to the

magnitude and timing of water-flow must be seen in the

hydrograph of a stream draining at least 0.1–50 km2. Similarly,

if impacts of a particular forestry practice on water-related

phenomena such as nutrient or sediment losses are to be

quantified, this must be observable at the 0.1–50 km2 scale.

Spatial variability is too great to attempt extrapolation of micro-

scale observations (<0.001 km2 or 0.1 ha) to the whole

landscape. Where changes are observed at the experimental

catchment scale, then local investigations may be needed to link

a particular forestry practice or forestry-related landform with

the catchment-scale impact. The extreme dearth of catchment-

scale studies addressing the hydrological impacts of logging

natural forests in the wet tropics (Chappell et al., 2004b) means

that: (a) extrapolation of results across the whole of the wet

tropics is unrealistic; and (b) more should be made of the

existing time-series data—at least to identify priorities for new

research studies.

1.2. Value of DBM models over models which have a priori

assumptions about the hydrological system being modelled

Modelling can be used to: (a) extrapolate time-series results

from case studies; and (b) identify key hydrological system

parameters or dynamic response characteristics. Three classes

of models that have been used to simulate the hydrological

response of rainforest catchments are: (1) physics-based

models; (2) conceptual models; and (3) Data Based Mechan-

istic models. Physics-based models of catchment hydrology

solve many geophysical equations for the component water-

paths within catchments, notably the Penman-Monteith

Equation for evapo-transpiration, the Chezy-Manning equation

for overland flow and the Darcy-Richards equation for

subsurface flow (e.g. Bathurst et al., 2004; Ott and Uhlenbrook,

2004). Controversially, we would suggest that no experimental

catchment (i.e. 0.1–50 km2) in the wet tropics has sufficient

data on soil/rock characteristics, soil depths, soil surface

characteristics and canopy characteristics, to allow accurate

physics-based modelling of distributed water pathways to be

made (see e.g. Beven, 2000, 2001b, 2002; Chappell et al., 1998,

2004c; Chappell and Sherlock, 2005; Croke et al., 2004).

Without reliable estimates of parameter values representative of

the whole catchment, forest use/change predictions become

unrealistic. Conceptual models make assumptions about the

dominant hydrological processes operating and then use

simplified equations to capture the hydrological dynamics.

Such models include TOPMODEL (e.g. Chappell et al., 1998),

TOPOG (e.g. Vertessy and Elsenbeer, 1999) and most

applications of IHACRES (e.g. Croke et al., 2004). DBM

models are different to both physics-based and conceptual

models in that they make no a priori assumptions about the

nature of the hydrological system, for example, the topographic

control on soil moisture (Chappell et al., 2004b, this volume) or

the dominance of two flow pathways (i.e. fast and slow: Croke

et al., 2004). In contrast, DBM modelling identifies a range of

simple mathematical relationships in the form of transfer

functions (TFs) sometimes combined with simple non-linear

sub-models, uses objective statistical criteria to reject some of

these models and only after this, rejects further where they are

shown to be inconsistent with a physics-based interpretation

(Young, 2001; Young et al., 2004). To reiterate, while DBM is

limited by the range of mathematical relations typically

evaluated (e.g. only three non-linear components: Young,

2003), it does not have a priori hydrological assumptions, so

that all identified models could be rejected if considered

inconsistent with hydrological theory. DBM is also different

from other data-based techniques (e.g. Camarasa and Tilford,

2002; Hundecha and Bardossy, 2004), in that it uses a very

powerful method, the simplified recursive instrumental variable

(SRIV) algorithm (Young, 1985, 1999, 2001), to identify the

model structures and associated parameter values.

Over the past 20 years, many scientists have suggested that

only physics-based catchment models should be used to

simulate land cover/use change impacts, such as those

associated with forestry (see e.g. Bathurst et al., 2004; Ott

and Uhlenbrook, 2004). More recently, other scientists have,

however, suggested that these physics-based catchment models

may not be giving reliable results of hydrological change

following forestry or other land cover/use change scenarios due

largely to the large uncertainties in the calibrated parameter

values (Parkinson and Young, 1998; Beven, 2001a,b; Beven

and Freer, 2001; Chappell and Fowell, 2003; Chappell et al.,

2004c). Models requiring calibration of a smaller number of

parameter estimates thus giving smaller uncertainty have

started to be used to simulate land cover/use change scenarios.

These models, with their smaller numbers of parameters (called

‘parsimonious models’: after Box and Jenkins, 1970) include

the conceptual models of Whitehead et al. (1998), Viney and

Sivapalan (2001) and Croke et al. (2004). Thus far, the DBM

modelling approach, which is also described as parsimonious,

has not been used to simulate tropical forestry impacts on

runoff response.

1.3. Tropical forestry impacts on runoff response

A key potential impact of tropical forestry on stream

response is via water pathway changes resulting from the

presence of vehicle tracks and roads (Chappell et al., 2004b).

Within the tropical rainforests of Southeast Asia, the most

common method of transporting logs from tree stumps to log

landings is by ground skidding using tracked skidder vehicles

(Thang and Chappell, 2004). These vehicles locally compact

topsoil, reduce its permeability (Van der Plas and Bruijnzeel,

1993; Malmer, 1996) and can locally enhance the proportion of

infiltration-excess overland flow (Baharuddin, 1995; Douglas

et al., 1995; Chappell et al., 2004a,b). There are, however, so

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–8060

Fig. 1. The 0.44 km2 Baru rainforest catchment (58010N and 117848.750E)located on the north-eastern tip of Borneo Island in Southeast Asia. The

boundaries of the contributory areas (solid lines), the perennial streams (dashed

lines) and lorry haulage roads (dotted lines) are shown. Contributory areas are

labelled at their outlet and gauging location. Raingauge locations are shown

with the symbol ‘R’. River gauging stations on channels with perennial flows

are shown with the symbol ‘P’, and those on channels with ephemeral flows are

shown with the symbol ‘E’.

few reliable data on the proportion of net rainfall which

generates infiltration-excess overland flow on skidder vehicle

trails in tropical rainforests (Chappell et al., 2004b; Thang and

Chappell, 2004). Indeed, we are only aware of the runoff plot

data on infiltration-excess overland flow on skid trails produced

by Baharuddin (1995) in Peninsular Malaysia, and those studies

within the 4 km2 Sapat Kalisun, which includes the 0.44 km2

Baru catchment (Table 1; Chappell et al., 2004b; Thang and

Chappell, 2004). Given that the Baru catchment and its

surrounding area might be considered a reference site for rates

of surface flow on tropical rainforest skid trails, more should be

made of these available data.

We know what the combined response of surface and

subsurface flows have on the response of the third-order Baru

stream under the given forestry regime, as we have

continuously gauged the flow of the stream at station P1

(Fig. 1: Chappell et al., 2004a). What we do not know, is how

the overall stream response would have differed if a different

density of skid trails (or other forestry features: Conway, 1982)

had been used within the Baru. Different densities of skid trails

are utilised with different forms of rainforest harvesting

practiced in Malaysia, notably clear-felling, conventional

selective logging and reduced impact logging (Pinard et al.,

1995, 2000; Thang and Chappell, 2004). To our knowledge,

there are no published modelling studies which show the

individual effect of skid trail density on stream response within

the wet tropics. The only study addressing the combined effects

of a different skid trail density and a different degree of canopy

disturbance for different types of selective rainforest logging in

the wet tropics is the Bukit Berembun study in Peninsular

Malaysia (Chappell et al., 2004b). The Water Erosion

Prediction Project (WEPP) model has been used to examine

the effects of road (and skid trail?) construction on surface and

subsurface flow pathways on streamflow response, but for

streams in temperate USA (Elliot and Tysdal, 1999; Rhee et al.,

2004). We chose not to use this physics-based modelling

approach to address possible skid trail impacts on streamflow

response for our tropical catchment because of the issue of

parametric uncertainty identified earlier. Instead, we will

Table 1

Plot studies in the 4 km2 Sapat Kalisun catchment (58010N, 117848.750E) nearthe Danum Valley Field Station (Malaysian Borneo), used to measure infiltra-

tion-excess overland flow on skid trails and natural slopes (disturbed and

undisturbed)

Slope type No. of plot

replicates

Type of

monitoring

Reference

Skid trail 1 Volumetric Sinun et al. (1992)

Disturbed cover 3

Undisturbed 3

Skid trail 1 Volumetric Douglas et al. (1995)

Haulage road

drainage

1 Datalogged

tipping-buckets

Chappell et al. (1999,

2004a), this study

Skid trail 1

Disturbed cover 3

Undisturbed 2

The 0.44 km2 Baru catchment is located in the northern headwaters of the Sapat

Kalisun.

attempt the first DBM approach to modelling possible impacts

of using different densities of forestry skid trails.

1.4. Aims and objectives

The aims of this work are to demonstrate the hydrological

information that can be extracted from high quality time-series

data (i.e. the value of collecting such data) at the catchment and

local scales, the hydrological interpretation of dynamic

response characteristics (DRCs) of different hydrological

pathways, and the utility of existing time-series data in

predicting the hydrological effects of tropical forest change, all

using DBMmodelling. The four specific objectives designed to

meet these aims were:

(1) T

o identify the relationship between catchment-average

rainfall and streamflow for a 0.44 km2 humid rainforest

catchment and the associated parameters (i.e. DRCs).

(2) T

o identify the possible structure and parameters of

component hydrological pathways (i.e. infiltration-excess

overland flow, subsurface flow and transpiration) of the

same catchment through DBMmodelling of the streamflow

(and subsurface flow) signal, and the first DBM modelling

of representative local-scale measurements of overland flow

and transpiration.

(3) T

o combine the component DBM models into a single,

multiple pathway model, that is consistent with observations

and the catchment-average rainfall–streamflow model.

(4) T

o use simple gain terms representing a long-term,

measured or predicted runoff coefficient for each water

pathway (at a representative local scale) to assess the effect

of differing skid trail density on streamflow via changes to

the extent of infiltration-excess overland flow.

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–80 61

The importance of fully utilising high quality, hydrological

time-series data to understand the behaviour of our managed,

equatorial catchment is stressed throughout. Indeed, this is the

first study to combine DBM modelling of rainfall–streamflow

time-series with representative local scale hydrological time-

series, and then explicitly utilise the results in scenario

modelling of humid tropical forestry.

2. Research catchment and time-series data available

2.1. Equatorial rainforest site and forestry management

The modelling presented here is based on time-series data

and hydrological experience gained from the 0.44 km2 Baru

Experimental Catchment in Sabah, Malaysian Borneo (Fig. 1,

58010N, 117848.750E: see Sinun et al., 1992; Chappell et al.,

1999). The catchment is less than 2 km northeast of the Danum

Valley Field Centre (DVFC), a research station managed by the

Research and Development Division of Yayasan Sabah, in

collaboration with research organisations such as the Universiti

Malaysia Sabah and the Royal Society of London. Yayasan

Sabah (YS) is a state forestry organisation that manages a

9728 km2 timber concession which is some 14% of the East

Malaysian State of Sabah. Forestry research within this region

has addressed the effects conventional tractor logging on

sediment delivery (Douglas et al., 1992), the effects of reduced

impact logging (RIL) on carbon budgets (Pinard et al., 1995),

the viability of enrichment planting in disturbed natural forest

(Moura-Costa et al., 1996), and the effects of post-logging

recovery on sediment delivery and canopy hydrology (Chappell

et al., 2001, 2004a). Environmental conservation is also a major

objective of YS. This is shown by their gazetting of three large

tracts of undisturbed natural forest as protected conservation

areas; these are the 438 km2 Danum Valley Conservation Area,

the 588.4 km2 Maliau Basin Conservation Area and ca.

300 km2 Imbak Canyon Conservation Area.

The Baru Experimental Catchment (Fig. 1) was subject to

the first phase of commercial selective logging in 1988/1989.

The history of logging operations in the Baru is shown in

Table 2. A combination of D8 tractor skidders and highlead

yarding was undertaken, and a yield of 79.9 m3/ha of timber

Table 2

History of the first phase of logging operations in the Baru Experimental Catchme

Date Forestr

Prior to August 1988 Gazett

timber

August–September 1988 Second

culvert

28th December 1988 Start o

during

10–18th May 1989 Harves

21st May 1989 Skidde

25th May 1989 Second

30th May–2nd Jun 1989 Highle

June 1989 Furthe

July 1989–to date Forest

researc

was harvested from the encompassing annual coupe (Moura-

Costa and Karolus, 1992). Given the dearth of hydrological

studies on selection felling, the Baru may have become the only

hydrological reference catchment in the wet tropics for such

forestry operations outside of those in Peninsular Malaysia

(Chappell et al., 2004b).

2.2. Hydrological time-series data available

Within this article we use time-series data on catchment

water-paths collected during the water year 1st July 1995 to

30th June 1996 (see Chappell et al., 1999), a period some 7-

years after road construction and timber harvesting of

previously virgin rainforest (Table 2). The time-series are

obtained by field sampling at 10 s intervals and modelling after

integration to 5 min intervals, giving 105,408 values per time-

series. Such a high resolution is required given the flashiness of

the streams in the Baru catchment (Chappell et al., 2004b).

The gross rainfall input to the Baru catchment was measured

with 103552D tipping-bucket raingauges (Casella CEL Ltd.,

Kempston, UK) monitored by Newlog dataloggers (Technolog

Ltd., Wirksworth, UK). Because of the high spatial variability

of rainfall within the Baru catchment (Bidin and Chappell,

2003), catchment average rainfall (at each 5 min interval) was

derived by applying the Thiessen Polygon method to the data

from six gauges distributed across the Baru Catchment (R1–R6

in Fig. 1). Fig. 2a shows the time-series of catchment average

rainfall for the Baru basin.

A gauging structure had been built to monitor the streamflow

leaving thewhole Baru catchment (P1 in Fig. 1). It comprises of

a thin-plate, 1208 V-notch weir which maintains sub-critical

flow in an upstream stilling pool where a shaft encoder linked to

a Newlog datalogger recorded water-level. The rating from

water-level to streamflow was then derived from the relation-

ship with spot measurements of discharge using dilution

gauging (integration method) and current metering (mean-

section method). Fig. 2b shows the time-series of streamflow

monitored at station P1 in the Baru catchment for 1st July 1995

to 30th June 1996 period. The Baru catchment contains 14 sub-

catchment areas, gauging a range of landscape elements from

channel heads, slopes with infiltration-excess overland flow,

nt, Ulu Segama Forest Reserve, Sabah, East Malaysia in 1988 and 1989

y operation

ing of Ulu Segama Forest Reserve for commercial forestry and inventory of

trees within the undisturbed natural forest

ary road and feeder road construction, including slope cutting,

ing and surfacing

f Matahari clearance of trees from the road corridor (to aid road trafficability

logging), skidder yarding and highlead yarding at first mast

ting of timber in the eastern part of the Baru catchment

r logging plus installation of second highlead logging mast

highlead mast in place in the western part of the Baru catchment

ad logging at second mast

r skidder yarding until end of June

recovery phase, with forest biomass monitoring and designation as a

h area

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–8062

Fig. 2. Five-minute sampled time-series of water flow for the 1st July 1995 to

30th June 1996 water-year in the Baru catchment, where (a) is the catchment-

average gross rainfall; (b) streamflowmeasured at station P1 (see Fig. 1); and (c)

the infiltration-excess overland flow measured at site E8 (see Fig. 1).

skid trails, to road drains and road gullies (Fig. 1). The larger

sub-catchment areas are similarly monitored with 1208V-notchweirs, while the smaller sub-catchment areas are monitored by

using zinc-plates and a flume to guide surface flow in to large

(ca. 3 l) tipping-bucket devices linked to Newlog dataloggers.

Sub-catchment areas where infiltration-excess overland flow

was monitored, were: (a) a lightly compacted, skidder-vehicle

track (E4 in Fig. 1) draining an area of approximately 0.155 ha;

(b) a section of haulage road that carries water that it intercepts

from a stream (E1 in Fig. 1), (c) a slope section that carries

surface flow when a road drain (E2 in Fig. 1) which intercepts a

stream (from site E9) that overtops and flows downslope

towards stream P4/5, (d) an ephemeral channel that receives

surface flow from a skid trail during storms (E5 in Fig. 1), and

(e) an area with an almost planar slope (lacking a channel, skid

trail or road) that generates some infiltration-excess overland

flow (E8 in Fig. 1). Fig. 2c shows the time-series of infiltration-

excess overland flow monitored at site E8 from 1st July 1995 to

30th June 1996.

The location of the timber haulage roads generating overland

flow within the Baru have been mapped by the Danum Survey

Team using closed-loop traverses with a TC400 Total Station

(Leica Geosystems AG, Heerbrugg, Switzerland; Fig. 1). These

data are supplemented by: (a) a detailed survey of the skid trails

within a 100 m wide corridor along 860 m of forestry road in the

eastern half of the Baru catchment; and (b) a survey of road and

skid trail width at 60 locations within the Baru.

The components of evapo-transpiration have been less well

monitored in the Baru catchment in comparison to rainfall,

streamflow, overland flow and a derived subsurface flow

estimate. The only 5-minute transpiration data currently

available for the Baru Experimental Catchment have been

measured on five 15-year old camphor (Dryobalanops

lanceolata) trees. These measurements were undertaken with

TDP80 sapflow sensors (Delta-T Devices Ltd, Cambridge,

UK), and together with air temperature, have been recorded

using CR10 dataloggers (Campbell Scientific Ltd., Shepshed,

UK). We have monitored rainfall below the forest canopy (i.e.

net rainfall) in a 4 km2 area including the Baru catchment using

500 storage raingauges and 5 tipping-bucket raingauges (Bidin,

2001; Bidin and Chappell, 2006). The 12-month average net

rainfall estimated as a proportion of the rainfall received by the

canopy (i.e. gross rainfall) has been calculated (Chappell et al.,

2001) allowing a 5-minute time-series of net rainfall to be

derived by multiplying the gross rainfall by this proportion. We

could not calculate a 5-minute time-series of wet-canopy

evaporation (i.e. that component of rainfall which evaporates

from wetted leaves rather than reaching the ground) as we do

not have measurements of canopy wetness (or canopy storage)

over the several hours that the leaves remain wet after rainfall

(cf. Bidin, 2001).

All of the time-series data pre-processing and then DBM

modelling was undertaken within the Matlab programming

environment (The MathWorks Inc., Natick, USA).

3. Methods

3.1. Method rationale and justification

All of the analysis used in this study used the Data-Based

Mechanistic (DBM) modelling approach (Young, 2001).

Within this approach the information contained within the

dynamics of lumped input–output behaviour (e.g. rainfall–

streamflow, or rainfall–overland flow and rainfall–subsurface

flow) was used to define the structure and parameters of the

model from a large range of possible options. Thus, a priori

hydrological information is not required in the initial stage of

modelling. This approach is consequently less constrained by

possible misconceptions about the hydrological theory (see e.g.

Klemes, 2000; Chappell, 2005) or the role of particular

phenomenon (e.g. soil layers in generating shallow lateral flow,

or an assumed role of Hortonian overland flow: Chappell and

Ternan, 1992; Chappell et al., 1999; Sherlock et al., 2000;

Chappell and Sherlock, 2005) than physics-based or conceptual

models. As DBM model structures are much simpler or more

‘parsimonious’ than those of physics-based models, uncertain-

ties in model predictions can be much reduced, and often more

readily calculated (Parkinson and Young, 1998). DBM and

some other data-based models (e.g. Croke et al., 2004) utilise

transfer functions (TFs) to simulate runoff or water pathway

dynamics. A first order TF, which has already been utilised to

simulate the rainfall to streamflow relation of the Baru

catchment (Chappell et al., 1999) takes the form:

QðtÞ ¼ P

1�Rz�1Reffðt � dÞ (1)

where Q(t) is subsurface flow along the dominant pathway (mm)

at time-step t; P the model production parameter; the recession

term; Reff the effective rainfall (i.e. rainfall corrected for non-

linear soil-water storage effects); d the pure time delay between

the effective rainfall input and initial streamflow response;

and z�1 the backward shift operator (Chappell et al., 1999).

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–80 63

TheparametersP and canbe used to derive the dynamic response

characteristics (DRCs) of:

ssP ¼ P

1�R(2)

TC ¼ �tbaselogeð�RÞ (3)

where ssP is the steady state production (or ‘water-balance’ term)

and TC is the time constant or ‘residence time’, and tbase is the

time base of the sampling (here 5 min) (Chappell et al., 1999).

Clearly, the robustness of any model structure and its

parameters is dependent on the sophistication of the model

identification/calibration routines (Box and Jenkins, 1970).

There are two key sophisticated elements in model identifica-

tion unique to DBM (andmore recent applications of the related

IHACRES), potentially making DBM more reliable than many

other modelling techniques:

(1) M

any models use least squares (LS) algorithms to find or

calibrate model parameters. The problem with these LS

methods is that the noise within time-series data gives rise to

considerable bias and uncertainty in the model parameters

(Young, 1984).With the DBMapproach, model structure and

parameter values are identified (or calibrated) by the

simplified recursive instrumental variable (SRIV) algorithm.

In the SRIVapproach, time-series data are pre-filtered using

the estimate of the dynamic recession filter after the initial LS

estimate. This removes higher frequency observation dis-

turbances focusing the signal in the useful frequency range

where dominant dynamic modes of the system are contained.

Following this step, the Instrumental Variable method with

specially generated instruments is used, which removes the

bias of the estimates. Both of these steps considerably reduce

parametric uncertainty (Young, 1984, 1985, 1999, 2001).

(2) A

second advantage of DBM over other data-based methods

in particular comes from its use of the objective statistical

measure of the Young Information Criterion (YIC) in the

model and parameter identification/calibration:

YIC ¼ loges2error

s2obs

þ logefNEVNg (4)

The first term of YIC is a measure of the model efficiency,

where s2error is the variance in themodel residuals and s2

obs the

variance in the observed data, and the second term, normal-

ised error variance norm (NEVN), is a measure of the degree

of over-parameterisation (Young, 1985, 2001). Generally, as

model complexity increases, so does the ability to capture

more and more of the dynamics in the observed time-series,

but at the expense of growing parametric uncertainty. There

is, therefore, an optimum complexity (or ‘model order’)

given that more complexity increases the uncertainty in

the individual parameters identified. The YIC is a measure

of whether the model has become over complex (i.e. too

many parameters) given the amount of information contained

within the observed data-series.

As a result, we consider that the DBM approach to

modelling does not give more combinations of water pathways

than is justified by the data, as can be the case with physics-

based and conceptual models which require definition of water

pathways prior to any modelling. The consequence of defining

the most simple structure justifiable and by using a method

which implicitly derives uncertainty estimates, DBM identifies

parameters (or DRCs) of the pathways that have well defined

values with quantified, small uncertainties.

3.2. Modelling stage I: DBM rainfall–streamflow response

The streamflow behaviour of a region is an integral of the

interacting effects of several water pathways within a catchment,

as such it is perhaps the most important hydrological time-series

to measure and understand within a tropical rainforest (Chappell

et al., 2004b). Thefirst stage of this study, therefore, usedDBM to

identify the dominant structure and parameter values for the

rainfall–streamflow response for the whole Baru catchment.

It is widely acknowledged that streamflow response is non-

linearly related to rainfall input. The most widely accepted

reason for non-linearity in catchment response is non-linear

variations in the moisture storage within the subsurface

unsaturated zone (Young and Beven, 1994; Chappell et al.,

2004b). Within the DBM methodology, this non-linearity is

typically modelled either using the Bedford Ouse Sub-Model

(BOSM) or the Store Surrogate Sub-Model (SSSM) (Chappell

et al., 1999, 2004a,b; Young, 2001, 2003). In this process the

rainfall input is said to be transformed to an effective rainfall,

Reff, which is then linearly related to the streamflow response.

For the analysis presented here, the Bedford Ouse Sub-Model

(Chappell et al., 2004b; Young, 2001) is used. The generalised

representation of BOSM for catchment response is:

ReffðtÞ ¼ RðtÞ ucatðtÞ (5)

where

ucatðtÞ ¼ ucatðt � 1Þ þ 1

tcatfRðtÞ � ucatðt � 1Þg (6)

where Reff is the effective rainfall (mm); R the gross rainfall

(mm); ucat(t � 1) the unsaturated zone storage variable at the

previous time-step (mm); and tcat the dimensionless non-lin-

earity term for the whole catchment response (Chappell et al.,

2004c). The non-linearity term (tcat) is determined by an

iterative process applied to the combined BOSM and TF

sub-models, where the objective function is the Nash and

Sutcliffe (1970) efficiency measure, R2t (Chappell et al.,

1999). The unsaturated zone storage variable (ucat) is calculatedby setting the first value to zero. The time-series of effective

rainfall is then normalised to give an ssP (Eq. (2)) equal to the

runoff coefficient for flow pathway (e.g. Eq. (9)) using:

RNeffðtÞ ¼ ReffðtÞ�

hPReffðtÞ

�(7)

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–8064

where h is a normalisation term, thus giving a normalised

effective rainfall, RNeff (mm). Thus, the non-linear form of the

DBM model of catchment response has five DRCs or para-

meters for a first-order model of the response, namely tu, h, TC(or ), ssP (or P) and d. Only five model parameters is con-

siderably less than 1000’s of parameter values typically

required by distributed, physics-based models, and conse-

quently such DBM parameter sets are much better defined

(Young, 2001; Chappell et al., 2004c).

The net rainfall is that proportion of the gross rainfall which

reaches the ground by stemflow and (direct or indirect)

throughfall (see e.g. Chappell et al., 2001). That component of

the gross rainfall which does not penetrate the vegetation

canopy, is stored and returned to the atmosphere by wet canopy

evaporation. Within the Baru catchment, the lag between gross

rainfall and net rainfall (due to temporary canopy storage) is

typically less than 5 min (Bidin, 2001), i.e. less than one time

step in the time-series data used within this paper, and thus not

significant. From the large raingauge network of Bidin et al.

(2003) that was monitored over one year, the net rainfall

amounts to 86% of the gross rainfall. Thus, the net rainfall time-

series is estimated from the formula:

RnetðtÞ ¼ aRgrossðtÞ; where a ¼ 0:86 (8)

where Rnet is the net rainfall below the vegetation canopy at each

5 min time-step (mm); Rgross the gross rainfall above the vegeta-

tion or in large forest gaps (Chappell et al., 2001) at each 5 min

time-step (mm), anda a gain termused to estimate the net rainfall

from the gross rainfall. DBMmodels of the relationship between

the catchment-average, net rainfall and streamflow were under-

taken, as with the rainfall–streamflow modelling.

For this first stage of the modelling process, only the

streamflow time-series and either the catchment-average

rainfall or catchment-average net rainfall time-series are

required, no additional information from local-scale measure-

ments or other catchment-scale measurements are needed. The

order of the model structure, for example, how many different

water pathways clearly express themselves in the lumped

rainfall–streamflow response was determined; with all possible

combinations from first-order to fourth-order models being

identified. While use of SRIV and YIC ensures that the model

structure and parameters are robust, it is difficult to know how

to modify the parameters of the model identified (i.e. tu, ssP, TCand d) to simulate the effects of a changing local or component

phenomena resulting from a different degree of tropical forestry

impacts. Thus, within the five subsequent modelling stages we

seek to model, concurrently, the component phenomena.

3.3. Modelling stage II: DBM net rainfall–local overland

flow response

At Danum, the proportion of the streamflow that is derived

by infiltration-excess overland flow or ‘Hortonian Overland

Flow’ (HOF: Horton, 1933), i.e. water which flows over the

ground surface due to the rainfall intensity exceeding the

infiltration capacity (or topsoil permeability) is less than 3% of

the rainfall on average (Chappell et al., 1999, 2004b). This is

because the local topsoil has a high permeability (Chappell and

Sherlock, 2005), while the typical storm rainfall intensity is

relatively low compared to tropical regions with cyclonic

rainfall (Chappell et al., 2001; Bidin and Chappell, 2003, 2006).

While it is commonly believed the tropical catchments have a

high proportion of infiltration-excess overland flow, field

measurements within most areas of the tropics (notably areas

without extreme cyclonic rainfall) show that little of the

streamflow is generated by infiltration-excess overland flow

(Dubreuil, 1985; Bonell, 2004; Chappell et al., 2004b; Bidin

and Chappell, 2006; Chappell, 2005).

Given that infiltration-excess overland flow generates such a

small proportion of streamflow, it is likely that its contribution

cannot be assessed by SRIV and subsequent partial fraction

expansion of the DBM-modelled, stream hydrograph (Young,

1992). For the Baru catchment we do, however, have direct

measurements of infiltration-excess overland flow, but at local

scales. For this preliminary study, all of the time-series of

infiltration-excess overland flow are examined and one selected

as being representative of the time-series of infiltration-excess

overland flow for the whole catchment.

Non-linearity may be apparent in the relationship between

net rainfall and infiltration-excess overland flow, due to the

variable effect of surface wetness and capillarity (Horton,

1933). Thus, we use the BOSM to examine the strength of the

non-linearity in the infiltration-excess overland flow process.

The form of the model follows that of Eqs. (5) and (6), except

that Reff and R are replaced by ROFeffnet (the effective net rainfall

that generates an overland flow response) and Rnet (the net

rainfall), respectively, while the non-linearity terms now relate

to the wetness of the ground surface, and are given as usur(t � 1)

and tsur. The effective net rainfall generating subsurface flow is

then normalised to give the same monthly input as the net

rainfall, as in Eq. (7), giving a normalised effective net rainfall,

RNOFeffnet (mm). The order and parameters of the linear transfer

function between the effective net rainfall and resultant

infiltration-excess overland flow are then identified or

calibrated with SRIV.

3.4. Modelling stage III: DBM net rainfall–subsurface flow

response

As only 4% of the streamflow of the third-order Baru

catchment is generated by overland flow, with 96% coming

from subsurface flow pathways (Chappell et al., 1998, 1999,

2004b,c), the non-linearity (tu) and residence time (TC)

parameters in the rainfall–streamflow model will be largely a

function of the subsurface flow behaviour. A subsurface flow

time-series can be calculated by subtracting a representative

infiltration-excess overland flow time-series from the time-

series of the Baru streamflow.

Within this study, the relationship between the subsurface

flow time-series and both the rainfall and net rainfall (see Eq.

(8)) is identified. As with the infiltration-excess overland flow,

the non-linearity is modelled using BOSM. The form of the

model follows that of Eqs. (5) and (6), except that Reff and R are

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–80 65

replaced by RSUBeffnet (effective net rainfall that generates a

subsurface response) and Rnet (the net rainfall), respectively,

while the non-linearity terms now relates to the subsurface

moisture, and given as usub(t � 1) and tsub. The normalised

effective net rainfall, RNSUBeffnet (mm) is then generated by the

same procedure as for the infiltration-excess overland flow or

whole catchment response (see Eq. (7)). The order and

parameters of the linear transfer function between the effective

net rainfall and resultant subsurface flow is then identified or

calibrated with SRIV.

3.5. Modelling stage IV: DBM temperature–local

transpiration response

During the water-year 1995/1996, the Baru catchment

received 2956 mm of rainfall and discharged 1867 mm of

streamflow. Thus, the runoff coefficient (RC) for the Baru

catchment, i.e.

RC ¼�P

QPR

�(9)

whereP

R is the annual catchment-average rainfall (mm) andPQ the annual streamflow leaving the same catchment (mm),

is approximately 0.63 (Chappell et al., 1999). With no major

aquifers within the catchment and an assumption of no sig-

nificant exchanges of groundwater across the catchment divides

(see Chappell et al., 1999), theP

P �P

Q value of 1089 mm,

or 37% of gross rainfall, equates to the evapo-transpiration.

Evapo-transpiration includes losses to the atmosphere by wet-

canopy evaporation, transpiration and soil evaporation. Within

a humid, well-vegetated rainforest environment, soil evapora-

tion is assumed to be insignificant. Wet-canopy evaporation in

the Baru has already been stated to be 14% of rainfall (i.e.

Pgross � Pnet over a year: Bidin et al., 2003). Thus, the annual

transpiration losses from the Baru catchment amount to

approximately 675 mm or 23% of gross rainfall in the 1995–

1996 water year (i.e. 37–14% of rainfall). This value is within

the range of those observed within other managed, equatorial

rainforests (Bruijnzeel, 1990; Kuraji, 1996) and is also quite a

significant component of the catchment water balance, so it

would be helpful to attempt to model its dynamics. We do not

have a high-resolution time-series of transpiration averaged

over the whole 0.44 km2 Baru catchment. Such data can be

derived from towers which extend well above the rainforest

canopy (Moreira et al., 1997), so we hope to utilise a 100 m

GAW tower recently constructed near to the Baru for these

measurements in the future. Presently, the only high-resolution

(5 min) data we have for transpiration comes from sapflow

sensors installed within individual camphor (D. lanceolata)

trees, all some 15-years old and 1 km southwest of the Baru

catchment.

In order to simulate the dynamics of the transpiration

pathway we sought to identify a new DBM model; air

temperature were used initially as the input variable. These

temperature data were manipulated in many ways before a

dynamic DBM model between a manipulated temperature and

transpiration time-series was identified. To allow the temporal

dynamics of transpiration from a single tree (ETRtree(t)) to be

used as an initial estimate of the catchment-average tran-

spiration dynamics (ETRC(t)), the tree data were normalised

using the difference in the annual average net rainfall and

streamflow, i.e.

ETRCðtÞ ¼ ETRtreeðtÞ�P

RnetðtÞ �P

QðtÞPETRtreeðtÞ

�(10)

3.6. Modelling stage V: linking components for subsequent

scenario modelling

Thus far, our methodology has used DBM techniques to

derive separate model structures and parameters for streamflow

generation and related water pathways. Such separate analyses

are important for understanding the dynamics of the whole

streamflow generation system, but a greater integration of these

component models might allow a new type of scenario

modelling, namely robust identification/calibration of compo-

nent DBM model structures from observed time-series,

combined with a simple ‘gain element’ to link the components,

which could be adjusted to reflect a forestry scenario. Within

this fifth stage of the modelling process, hydrological under-

standing of the linkages between the component water

pathways is essential. The key reason for linking these

components is to allow the balance between specific

components (e.g. infiltration-excess overland flow and subsur-

face flow) to be altered using hydrological experience from the

local or related sites. This could allow the effect of specific

forestry operations on the overall streamflow response to be

examined simply via modification of the balance between

different component pathways. This method has the very

clearly expressed assumption that DRCs of each component

pathway do not change with the land cover/use modification.

For example, flashiness (i.e. TC) of the overland flow pathway

or its sensitivity to surface moisture (i.e. tugs) does not change

as the number of skid trails in an area changes as a result of

different harvesting rates; only the amount of water moving

along this pathway changes. This is unlikely to be true, but may

be considered acceptable as a first approximation. This type of

scenario modelling could have several advantages over

scenario simulations with physics-based models. First, the

initial or unchanged state of the system is based directly on the

time-series data available for the catchment, rather than

assumed relationships between small-scale physics (and

associated parameters) and the water pathways (sometimes

not even observed) (see Beven, 2001b; Chappell et al., 1998,

2004c). Second, allowing different forestry scenarios to be

simulated by changes to the balance between different

hydrological components (i.e. via a gain term) allows time-

series observations of the relative importance of different

pathways from other studies to be incorporated simply and

clearly.

Within the example forestry scenario that follows, the key

linkage is between the DBMmodel of net rainfall to infiltration-

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–8066

excess overland flow and that of net rainfall to subsurface flow.

Once the two separate DBM models are derived from the

observations, two gain terms are added to alter the magnitude of

the input, one for each pathway:

RnetOF ¼ Rnetb (11)

RnetSUB ¼ Rnetð1� bÞ (12)

where

b ¼�fscenario

fobs

�(13)

fobs ¼�P

QOFPRnet

�(14)

and RnetOF is the component of net rainfall entering the non-

linear infiltration-excess overland flow system; RnetSUB the

component of net rainfall entering the non-linear subsurface

system; fobs the annual proportion of net rainfall observed to

generate infiltration-excess overland flow; fscenario the annual

proportion of net rainfall observed to generate infiltration-

excess overland flow set by the user to the observed value

(fobs) or an alternative proportion of infiltration-excess over-

land flow for a forestry scenario;P

QOF the annual total of

infiltration-excess overland flow (mm);P

Rnet the annual total

of net rainfall (mm); and b the scaling factor in the gain element

to allow for a different balance of overland and subsurface flow.

The overall model comprising of the individual DBMmodel

components linked by either: (i) a switch (between net rainfall

to streamflow or net rainfall to component surface and

subsurface pathways); (ii) a gain element (for infiltration-

Fig. 3. A block diagram of th

excess overland flow and subsurface flow); or (iii) an integrator

(for transpiration), we call the BARUMODEL, after the

catchment data-series used to develop the approach. A block

diagram of the BARUMODEL is shown in Fig. 3.

3.7. Modelling stage VI: skid trail impact scenarios

Our first application of the BARUMODEL structure is in the

assessment of the hydrological effects of a key tropical forestry

operation, namely ground skidding (see Conway, 1982). We

begin this work by reviewing the literature for the different

densities of skid trail used within the wet tropics, notably within

Malaysian rainforest. The second stage of the work attempts to

utilise this information within the BARUMODEL. For the

scenario modelling presented in this paper, we make

preliminary assumptions that the non-linearity and flashiness

of the overland flow system does not change under a different

forestry system (e.g. clearfell, conventional selective logging

and reduced impact logging) only the proportion of the water

moving along this surface pathway changes. We know that this

assumption is not fully valid, but need to make it before we can

evaluate it in future work. We utilise the skid trail density

information by simply changing the fscenario term and thereby

the gain element b-term, to alter the split between infiltration-

excess overland flow and subsurface flow.

4. Results: DBM model and parameter identification

4.1. DBM rainfall–streamflow response

The results of the DBMmodelling of the rainfall–streamflow

are shown in Table 3. To show the value or need for the non-

linear transform of the rainfall to take account of the

e whole BARUMODEL.

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–80 67

Table 3

DBMmodel parameters identified for water pathways lumped at the third order

Baru stream (58010N, 117848.750E) near the Danum Valley Field Station

(Malaysian Borneo)

Parameters and

statistics

DBM model

1 2 3

Structure [1 1 2] [1 1 3] [1 1 3]

R2t

0.488 0.769 0.769

YIC �11.141 �12.499 �12.499

t – 300 300

h – 1438 1237

�0.9678 �0.8739 �0.8739

s() 0.0002 0.0005 0.0005

P 0.0246 0.0794 0.0923

s(P) 0.0001 0.0003 0.0004

TC 152.55 37.10 37.10

s(TC) 1.00 0.17 0.17

ssP 0.7618 0.6299 0.7323

RC* 0.630 0.630 0.733

Model 1: purely linear DBMTFmodel of gross rainfall to streamflow; Model 2:

DBM TF with BOSM non-linear model of gross rainfall to streamflow; Model

3: DBM TF with BOSM non-linear model of net rainfall to streamflow.

Structure: [No. of denominators, numerators, pure time delays]; R2t : Nash

and Sutcliffe (1970) efficiency measure; YIC: Young Information Criterion

(see Eq. (4)); t: BOSM non-linearity term (see Eqs. (5) and (6)); h: normal-

isation term (see Eq. (7)); : recession term; s(): standard deviation of the

recession term; P: production term; s(P): standard deviation of the production

term; TC: time constant (or residence time); s(TC): standard deviation of the

time constant; ssP: steady-state production (or gain) of the transfer function

(TF); RC*: observed input minus output of the pathway being modelled.

unsaturated zone storage effects, a purely linear transfer

function model was first derived. This purely linear first-order

model explains only 48.8% of the variance in the streamflow

responsewith one P-value (hence one steady-state production)

and one -value (hence one time constant) and three time-steps

(i.e. 15 min) of pure time delay before the initial response (d)

(Table 3). By capturing the non-linearity usingBOSM (Eqs. (5)

and (6)), the first order model explained 76.9% of the variance

in the streamflow response (Table 3), a significant improve-

ment over the purely linear model. SRIV was used to calibrate

the DBM model to 12-months of 5 min sampled data (water

year 1st July 1995 to 30th June 1996). Given the time-series

data has a 5 min time-step, the -value of—0.8739 � 0.0005 for

the non-linear model equates to a residence time or time

constant (Eq. (3)) of approximately 37.10 � 0.17 min. Thus,

the uncertainty in this measure (given as one standard

deviation, see Table 3) of the flashiness of the hydrological

system is given. The uncertainty arises from a combination of

Table 4

Time-constants (TCs), or ‘residence times’, for first order DBM models several sm

Time constant Catchment Area (km2) TF model structure Non-

14.5 min Plot 0.000015 [1 1 0] SSSM

23 days C1 0.133 [1 1 0] SSSM

37 min Baru 0.44 [1 1 3] BOS

8.6 h Coalburn 1.5 [1 1 0] BOS

8.3 h Bottoms 10.6 [1 1 1] BOS

Note that all DBMmodels shown use a Store Surrogate Sub-Model (SSSM: Chappe

the initial non-linear response, but the differences in the TCs of the linear compon

noise in the rainfall and/or streamflow data (due to rainfall

patchiness, errors in the gauging structure calibration, etc.) and

limitations of the model structure to fully capture all of the

dynamics of the system. Further, a time constant of

37.10 � 0.17 min indicates a very flashy rainfall–streamflow

response in comparison to other contributory areas modelled

using DBM techniques (Table 4). A key comparison site is the

0.133 km2 Bukit Berembun C1 catchment which is also in

Malaysian rainforest and has a similar convective rainfall

regime to the Baru (Bidin, 2001; Noguchi et al., 2003; Bidin

and Chappell, 2003, 2006). The geology does, however,

contrast markedly between the two sites and is probably the

main reason for the difference in the flashiness (or time

constant) of the two basins. The Bukit Berembun catchment is

on a deeply weathered granite (i.e. saprolite) aquifer giving the

more damped response (i.e. greater storage), while the Baru is

one a relatively impermeable mudstone aquitard (Chappell

et al., 2004b).

Higher order modelling with up to fourth order DBMmodels

was undertaken. Many models converged, but showed too large

a positive change in the YIC measure (Eq. (4)) from that of the

first order model values or produced Impulse Response

Functions with non-monotonic recessions (see Box and

Jenkins, 1970). The first test is an indication of over-

parameterisation of the model structure, while the second test

is an indication of a model with oscillatory behaviour. All

higher order models were, therefore, rejected, leaving only the

first order model.

In order to show the effect of subtracting a fixed proportion

of wet-canopy evaporation from the rainfall (Eq. (8)), the net

rainfall to streamflow response was also modelled with the

DBM approach. Table 3 shows that the DRCs of the non-

linearity measure (tu) and flashiness measure () for a first-order

model do not change significantly from those of the gross

rainfall–streamflow response (Table 3). To illustrate how much

of the observed streamflow dynamics are captured, a short

period of observed and simulated data are presented in Fig. 4.

These data cover a storm with a return period of 10 years

(Douglas et al., 1999; Chappell et al., 2004a). This model used a

calibrated tu-value of 300 (Eqs. (5) and (6), Table 3) and is

shown in Fig. 5.

4.2. DBM net rainfall–local overland flow response

On all monitored slopes at Danum, except timber haulage

roads and skid trails, annual mean infiltration-excess overland

all contributory areas ranked by area

linear filter Climatic regime Media Reference

Temperate Acid soil Fawcett et al. (1997)

Humid tropical Saprolite Chappell et al. (2004b)

M Humid tropical Mudstone This study

M Temperate Mudstone Chappell, unpublished

M Temperate Limestone Chappell, unpublished

ll et al., 1999) or Bedford Ouse Sub-Model (BOSM: Eqs. (5) and (6)) to capture

ent resulting from using either model is very small.

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–8068

Fig. 4. Simulated DBM streamflow from net rainfall input together with

observed streamflow of the 0.44 km2 Baru catchment. The model is calibrated

to a 12 month time-series sampled every 5 min (105,408 values per time-series),

but shown for an example storm with a 10-year return period (i.e. 19th January

1996).

Fig. 5. The first order DBMmodel of net rainfall to streamflow for the 0.44 km2

Baru catchment for the 1st July 1995 to 30th June 1996 water year.

flow was less then 5% of the stream hydrograph (Sinun, 1991;

Sinun et al., 1992; Chappell et al., 1999, 2004a). Only on very

compacted sections of skid trails (made by the D8 skidders)

near to haulage roads, were significant volumes of infiltration-

excess overland flow generated (e.g. 52.5% at site 5: Sinun,

1991; Sinun et al., 1992). Indeed, on skid trails used for only a

few tractor passes, e.g. site E4 (Fig. 1, equivalent to site 4TA in

Chappell et al., 1999) infiltration-excess overland flowwas only

2.4% of net rainfall. Overland flow was monitored on two

haulage roads in the Baru catchment-site E2 and E1, and 4 and

62% of streamflow per unit area of catchment was generated

(Chappell et al., 1999, 2004a). The flow at site E2 was caused

by an upslope road drain over-toping, because the stream (E9

stream in Fig. 1) had not been culverted beneath this secondary

haulage road. Similarly, the flow at site E1 was caused by the E1

stream not being culverted beneath the feeder road, so that the

whole stream is diverted along the road. In both cases,

significant volumes of overland flow are not generated by a

local infiltration-excess mechanism, but by the roads inter-

cepting and re-routing channel flow along their surface. Indeed,

the length of the path taken by the channelized flow actually

increased following re-routing of the flow along the road

sections.

Given these varied results, the time-series of overland flow

for site E4 is judged to best represent the overland flow

dynamics for the whole Baru catchment, in this preliminary

study. Further work modelling all of the overland flow time-

series presented in Chappell et al. (2004a) is ongoing. By

applying the DBM approach to the E4 data, a first-order transfer

function is seen as the optimal model order (Table 5), as higher

order models do not have acceptable changes in YIC values.

The overall relation is seen to be non-linear (Table 5), though

less strongly than the overall rainfall–streamflow response. In

the case of the overland flow process, this is likely to be due to

the effect of variable near-surface soil-water content. Compar-

ison of the modelled infiltration-excess overland flow time-

series for site E4 with that observed is shown for an example

rainstorm with a return period of 2 months (ca. 50 mm: Douglas

et al., 1999; Bidin, 2001) in Fig. 6, though the whole water year

is modelled. The resultant surface flow model and its

parameters (based on time-series of 105,408 values) are shown

in Fig. 7 and Table 5. It can be seen from the estimate of the time

constant of 5.499 � 0.085 min, that the infiltration-excess

overland flow response is much flashier than the streamflow

response. This is expected given the limited surface storage and

hence less damping of response in comparison to a streamflow

response affected by subsurface moisture storage.

4.3. DBM net rainfall–subsurface flow response

Simulation of the subsurface flow was achieved by

subtracting the time-series of flows measured at site E4 (i.e.

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–80 69

Table 5

DBM model parameters identified for component water pathways in the third

order Baru stream (58010N, 117848.750E) near the Danum Valley Field Station

(Malaysian Borneo)

Parameters

and statistics

DBM model

1 2 3 4 5

Structure [1 1 1] [1 1 1] [1 1 3] [1 1 4] [1 1 0]

R2t

0.601 0.732 0.481 0.733 0.794

YIC �9.578 �9.492 �11.009 �12.124 �8.812

t – 385 – 320 –

h – 1415 – 1176 –

�0.5224 �0.4028 �0.9681 �0.8685 �0.9060

s() 0.0049 0.0057 0.0002249 0.0006262 0.0025

P 0.0225 0.0143 0.0263 0.0093 0.00059

s(P) 0.0002 0.0001 0.0002 0.0004 <0.00001

TC 7.70 5.499 154.20 35.46 50.7

s(TC) 0.11 0.085 1.10 0.18 1.40

ssP 0.0472 0.0240 0.8258 0.7088 0.0063

RC* 0.024 0.024 0.709 0.709 –

Model 1: purely linear DBMTFmodel of net rainfall to overland flow;Model 2:

DBM TF with BOSM non-linear model of net rainfall to overland flow; Model

3: purely linear DBM TF model of net rainfall to subsurface flow; Model 4:

DBMTFwith BOSMnon-linear model of net rainfall to subsurface flow;Model

5: purely linear DBM TF model of rising temperature to transpiration. See

Table 3 for definition of parameters.

infiltration-excess overland flow) from the streamflow response

(i.e. channel-flow measured at site P1 in Fig. 1). The results of

the DBMmodelling of the net rainfall–subsurface flow relation

are shown in Table 5. As with the net rainfall to streamflow

model (Table 3), a purely linear model explains only 48.1% of

the subsurface flow (Table 5). Similarly, inclusion of the BOSM

non-linear filter, here related to unsaturated zone moisture

dynamics, increases the level of explanation to 73.3% (Table 5).

This first-order model for the subsurface flow response has a

pure time delay of four time steps (ca. 20 min) and a time

constant (Eq. (3)) of 35.46 � 0.18 min. As the streamflow is

Fig. 6. Infiltration-excess overland flow modelling results for contributory area

E4 in the Baru catchment, showing the time-series of the modelled (solid line)

and observed (broken line) flow illustrated with an example storm with a 2

month return period (i.e. 28th September 1995), but calibrated to thewhole year.

largely derived from subsurface flows, it is not surprising that

this value is very similar to the 37.10 � 0.17 min TC for the net

rainfall–streamflow response. Again, such a rapid subsurface

pathway probably relates to the absence of a deep aquifer, and

hence restriction of flow to shallow soil layers (Chappell et al.,

1998, 1999, 2004b) and predominantly convective rainfall

regime (Bidin and Chappell, 2006), but also may relate to the

presence soil piping within the subsurface system (Chappell

et al., 1998; Chappell and Sherlock, 2005). As with the

streamflow response, a statistically acceptable higher-order

decomposition of the Baru net rainfall–subsurface flow

response was not observed.

4.4. DBM temperature–local transpiration response

A DBM model explaining much of the variance in

transpiration dynamics for individual trees was found after

Fig. 7. DBM model of infiltration-excess overland flow for site E4, Baru

catchment.

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–8070

manipulation of air temperature time-series and subsequent

transfer function modelling using the SRIValgorithm (Table 4).

The existence of this relationship is probably due to both air

temperature and transpiration being strongly controlled by the

net radiative energy available until midday when the stomata

close. For the initial estimate of the catchment-average

transpiration presented here, the model for camphor tree K4

scaled to catchment water balance (Eq. (10)) was derived. The

temperature time-series associatedwith this treewasmeasured at

1 m above ground, 0.5 m away from this tree (shownwith a solid

line in Fig. 8a). The method we developed first derived observed

transpiration rates (mm per 5 min time-step) from the sapflow

velocities monitored at tree K4 by multiplying by the sapwood

area. These rateswere then scaled by the annual proportion of net

rainfall generating the streamflow (i.e. 0.23: Eq. (10)), so that the

rates are more representative of the catchment-average

transpiration. Next the annual minimum temperature was

subtracted from the 5 min resolution time-series of air

temperature. Readings where temperature had fallen from the

Fig. 8. The model of transpiration from the Baru catchment shown over an

example 5-day period in the water year. Subplot: (a) shows the air temperature

(solid line) and rising-stage, air temperature (circles) and (b) shows the

measured sapflow from camphor (D. lanceolata) tree K4 scaled by the Baru

water-balance (dotted line) and the modelled transpiration (solid line).

previous time-step were then set to zero (mostly the afternoon,

when stomata are closed or closing) and the resultant time-series

(called EN(t)) derived and shown with open circles in Fig. 8a for

an example 6-day period. The SRIValgorithm was then used to

derive a linear DBM model of the transpiration.

The resultant DBM transfer function model of catchment

transpiration has a structure and parameters given as:

ETRCðtÞ ¼0:0010

1� 0:9060z�1ENðtÞ; (15)

ENðtÞ ¼ TriseðtÞ � Tmin (16)

where ETRC(t) is the catchment transpiration at time-step t (mm

per 5 min period), and EN(t) is the rising-stage, air temperature

at each time-step (Trise(t)) minus the annual minimum air

temperature (Tmin). Both temperature values are in 8C.Table 5 shows the model parameters and statistics identified.

This linear transfer function model using EN(t) was found to

able to explain 79% of the variance in the observed transpira-

tion data (Fig. 8b, Table 5), and helps us understand some of the

temporal dynamics of the transpiration processes within rain-

forest trees of the Baru catchment. First, we can see from the d-

value (Table 5), that there is no time delay between a tempera-

ture rise and a change in transpiration rate. Second, a time

constant of only 50.7 � 1.4 min (Table 5) indicates that the

transpiration pathway is a very responsive (limited storage)

system, when compared to, for example, subsurface pathways

in major aquifers (Table 4). We are continuing to evaluate this

transpiration model for other trees (with the same and different

model parameters) and are looking at the role of the controlling

variables of wind speed, saturation deficit of the atmosphere,

soil capillary potential and net radiation in DBM simulation of

transpiration.

4.5. Linking components for subsequent scenario

modelling

Each of the separate DBM models, with their structure and

parameters partly derived by the SRIV algorithm, were then

combined to give an output comparable to that simulated by the

lumped net rainfall to streamflow model and the observed data

(Fig. 9). While a streamflow time-series can be constructed by

combining the infiltration-excess overland flow and subsurface

flow models or simulated results, this second-order behaviour

could not be identified from the lumped rainfall–streamflow

model because the overland flow component is such a small

proportion of the streamflow (i.e. 4.4% of streamflow or 3.2%

of net rainfall). Gain elements were then added between the net

rainfall and the overland and subsurface flow models (Eqs. (11)

and (12)), to allow for changes in the amount of overland flow

generated as part of scenario modelling. For simulation of the

observed time-series, the gain element of the fscenario parameter

(and hence b-parameter) was derived from the observed data

(i.e. observed annual proportion of the net rainfall generating

overland flow) and thus had no impact on the simulated

responses. The ability of the user to manually change this

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–80 71

Fig. 9. Observed streamflow in the Baru catchment together with the stream-

flow estimated by adding the DBM overland flow and DBM subsurface flow

models. The model is derived from 12 months data, but illustrated with an

example storm with a 2 month return period (i.e. 28th September 1995). The

broken line shows the observed streamflow, the solid grey line shows the

simulated streamflow estimated by the lumped net rainfall–streamflow model,

while the fine solid line is from the combined surface flow (2.4% net rainfall)

and subsurface flow model. The simulated streamflow for the combined surface

flow (3.2% net rainfall) and subsurface flow model is not shown, but is very

similar to the combined model that is shown.

parameter does, however, allow for forestry scenario modelling

being undertaken.

The very simple model taking account of the losses due to

wet canopy evaporation is effectively a simple gain element

with a single parameter a; this is shown in Eq. (8) and Fig. 3.

With new data on canopy storage this static model could be

developed into a dynamic model. The transpiration model is

linked to the other water pathways within the BARUMODEL

using an integrator (Fig. 3) that normalises the gain to the

annual water-balance. Again, with more data on the controls of

transpiration, for example, the effect of canopy wetting on

transpiration suppression, a more dynamic linkage with the

other BARUMODEL components could be developed.

5. Results: preliminary application to skid trail impacts

Within the first version of the BARUMODEL with its linked

DBM-derived components, the surface–subsurface partition (b-

value) was changed to simulate two forestry scenarios. Our

objective was to focus on the effects of a different density of

skidder vehicle trails (or ‘skid trails’) on the partition between

surface and subsurface flow, if other types of forestry practice

(i.e. clearfelling or reduced impact logging) had been used

instead of conventional selective logging (see Thang, 1987;

Thang and Chappell, 2004).

5.1. Appropriate h and fobs values for BARUMODEL

To define appropriate values of h and fobs for BARUMO-

DEL (see Eqs. (7) and (14) and Fig. 3), it is first necessary to

review the measured extent of tropical forestry roads and skid

trails. Several studies in the wet tropics (e.g. Sinun et al., 1992;

Baharuddin, 1995; Douglas et al., 1995; Chappell et al., 1999,

2004a) have shown that localised sections of forest road can

generate significantly more infiltration-excess or Hortonian

overland flow (HOF) in comparison to adjacent slopes. Similar

findings have been obtained for rural roads in the seasonal

tropics (Ziegler et al., 2001). The main question is whether

localised changes along skid trails associated with tropical

forestry have a significant impact on the behaviour of micro-

catchment (0.1–1 km2) areas (Chappell et al., 2004b). Local

HOF, because of variation in road orientation and road micro-

topography, may run on to side slopes at many locations where

it then infiltrates, rather than connect directly to a stream (Sidle

et al., 2004; Chappell, personal observations in the Baru

catchment, 1990–2004). As a result of these two factors (road

orientation and micro-topography), the presence of HOF along

short sections of even the most compacted roads might not have

a significant impact on stream hydrographs. A noteworthy

study undertaken in the north-western USA (Jones and Grant,

1996) does, however, suggest that forest haulage road

construction can affect stream hydrographs, notably by

increasing storm peakflows. Even with this American study,

others (e.g. Thomas and Megahan, 1998) have suggested that

the results do no apply to large storms. A major problem for the

identification of forest road or skid trail impacts on stream

hydrographs in the wet tropics is the lack of high quality

streamflow records prior to road/trail construction (Chappell

et al., 2004b). This means that modelling possible changes to

stream hydrographs resulting from different skid trail scenarios

may be useful, at least to help define the scale, duration and

local conditions in which to undertake the new field studies

required.

The Baru catchment was subject to the first phase of

commercial selective logging in 1988/1989 (Table 2). The

period of the example analyses (i.e. 1995–1996) is 7-years post-

logging activity, so some vegetation recovery on less

compacted skid trails will have taken place (Douglas et al.,

1995), thus potentially affecting hydrograph shape. Within

Malaysian forest reserves, several two-lane, surfaced roads for

high speed log transport to the saw mill are constructed-these

are called ‘primary haulage roads’ (Forestry Department

Peninsular Malaysia, 1999). These main roads connect to

‘secondary haulage roads’ that are constructed to serve two or

three annual logging coupes. These roads are also surfaced, but

normally only have a single lane. Within individual annual

logging coupes, un-surfaced ‘feeder roads’ allow timber lorry

access to more distant areas. Within the Baru catchment, we

calculate that a total of 1094 m of ‘secondary haulage road’ was

constructed just below the northern divide (Fig. 1), and this had

been surfaced with a 0.2 m depth of local chert. A further 719 m

of un-surfaced, feeder road was located on the western

catchment divide (Fig. 1). The lorries that used these roads,

collected the timber from ‘log landing areas’.

In most rainforest areas of SE Asia, timber is brought to the

log landing areas by cableways (e.g. highlead: Conway, 1982)

or skidders, only occasionally are helicopters used. Both of

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–8072

these main primary transportation systems were used within the

Baru catchment; the area containing two highlead masts and

many skid trails. A recent survey of a 100 m wide corridor

along 860 m of the eastern section of the secondary haulage

road in the Baru showed 528 m of skidder vehicle trails are

present. Thus, for a unit area of the Baru catchment, we

calculate that there are 25 m/ha of secondary road, 16 m/ha of

feeder road and 61 m/ha of skid trails (Table 6). These forest

road/trail surveys show that within the Baru, road/trail densities

comply with the Standards of Performance (SOP) required in

those ‘lowland Dipterocarp forests’ that are certified by

organisations such as SGS as sustainably managed (Table 1,

Thang and Chappell, 2004). The width of the compacted

surface of the secondary road in the Baru was found to be

approximately 4 m, while the feeder roads and skid trails were

approximately 3 m. Thus, the area of the Baru covered by

secondary road is approximately 4,376 m2 or 1.0% of whole

catchment, by the feeder road 2,157 m2 or 0.5%, and by the skid

trails 8103 m2 or 1.8% (if the measured corridor area is

representative). Thus, all of the road/trail surfaces occupy only

3.3% of the whole Baru.

To define appropriate values of h and fobs for BARUMO-

DEL (see Eqs. (7) and (14) and Fig. 3), it is secondly necessary

to estimate the proportion of infiltration-excess overland flow

from different tropical rainforest surfaces. Sinun (1991) and

Sinun et al. (1992) reported a mean proportion of infiltration-

excess overland flow generated by a mix of disturbed and

undisturbed slope sections which lack roads or trails of 2.5% of

net rainfall. This figure is similar to the 2.8% of gross rainfall

reported by Malmer (1996) on similar soils in managed

rainforests in north-western Sabah. At Danum, the highest

recorded proportion of infiltration-excess overland flow

generated by a skid trail surface was the 52.5% of gross

rainfall recorded by Sinun et al. (1992). All of the

measurements of Sinun et al. (1992) were undertaken in

1989/1990, shortly after the logging activity in the Baru

(Table 2). Most skid trails and un-surfaced feeder roads within

the Baru have developed a vegetation cover in the 7-years post

logging activity. The feeder road in the Baru is covered by

Table 6

Density of haulage roads, feeder roads and skid trails within selected catchments

Peninsular Malaysia

Site Logging operation Density

Haulage

Actual road and trail densities in selected Malaysian rainforests

(1) Baru Selective 25

(2) Bukit Berembun Selective (CONV) 0

(3) Bukit Berembun Selective (RIL) 0

(4) Bukit Tarek (C3) Clearfelling 0

Current maximum allowable road and trail densities (standards of performance)

(5) Peninsular Malaysia Selective –

(6) Sabah Selective –

Notes: (1) FromChappell (unpublished); (2) and (3) derived from amap in Abdul Rah

logging, also called reduced impact logging (RIL); (4) fromNoguchi et al. (2004); (5)

a survey of skid trails within an 860 m long and 100 m wide corridor centred on the

estimate for the catchment as a whole).

grasses, vines and shrubs, while most skid trails are covered by

shrubs and small trees. Douglas et al. (1995) demonstrated that

such vegetation re-growth significantly reduces the proportion

of HOF generated. In some contrast, the skid trail section that

generated the 52.5%HOF remains only sparsely vegetated due

in part to the very high degree of compaction (see, Nussbaum

et al., 1995). Thus, the 52.5% HOF may be considered

representative of some sections of well-used feeder roads and

skid trails, but not representative of most skid trails in the Baru,

with their lower levels of compaction. Indeed, some skid trail

sections in the Baru (e.g. site E4 in Fig. 1) generated only 2.4%

of net rainfall as HOF some 7-years post logging (Chappell

et al., 1999). Critically, even where skid trails or roads generate

local HOF, a significant quantity is seen to run-off onto side

slopes where it then infiltrates rather than directly enter the

stream channels (Chappell, personal observations in the Baru

catchment, 1994–2004). As a consequence of these two

factors, reducing the observed proportion of HOF along skid

trails from the measured highest 52.5% value (on a very

compacted surface) to say 20% would not be unrealistic for a

typical road/trial surface. Overland flow was not visible on the

secondary haulage road in the Baru, except where streamflow

from an upslope channel discharged onto the road during

storms (e.g. site E2 in Fig. 1). The high infiltration capacity of

the coarse chert surface of the haulage road allows rainfall to

infiltrate. Some of this water may: (a) percolate in an upslope

direction to supply the road drain (a proportion of which then

inputs directly to a stream channel), but the remaining

proportion probably (b) percolates into the downslope soils.

Thus, the 20% HOF value for the skid trails is not unrealistic

for the secondary roads also.

If these very approximate estimates from the three road/trail

surfaces are combined with the figure for the other slopes, then

a value for the HOF proportion for the whole Baru catchment

would become:

ð0:04 fractional area� 20%Þ þ ð0:96 fractional area� 2:5%Þ¼ 3:2% of net rainfall (17)

, together with the maximum values currently allowed in certified forests in

(m/ha)

road Feeder road Skid trail Total

16 61* 102

56 69 125

60 22 82

52 176 228

<40 <300 –

<20 – –

im (1990) for conventional selective logging (CONV) and for closely supervised

and (6) see Thang and Chappell (2004) for lowland Dipterocarp forest. (*) From

secondary haulage road (given the proximity to the road, this possibly an over-

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–80 73

Even if the highest measured HOF value (i.e. 52.5%) was used

to represent all road and trail surfaces, this would still only give

a value of 4.5% HOF for the Baru as a whole.

As a result of these analyses of road/trail contributions to

overland flow in the Baru catchment, the observed proportion of

net rainfall that becomes infiltration-excess overland flow (i.e.

0.024 at site E4), we scale to 3.2% of net rainfall. The result of

this scaling has only a relatively small impact on the model

parameters (Table 7) or hydrograph form. Fig. 9 shows the

observed streamflow for the example storm with a return period

of 2 months with the time-series constructed by combining the

3.2% infiltration-excess overland flow and associated subsur-

face flow modelling results. For the whole 12-month period, the

combined model with 3.2% overland flow explains 76.67% of

the variance in the observed streamflow, which is only slightly

less than the combined model with 2.4% overland flow

(76.99%) or the lumped net rainfall–streamflow model with

77.43% of variance explained (Table 3). Thus, the combined

surface–subsurface models would be only slightly less accurate

for forecasting streamflow in comparison to the lumped model,

but they have the advantage that the effects of modifying the

surface–subsurface partition can be readily undertaken and

seen.

5.2. Appropriate fscenario and b-values for forestry skid

trail scenarios

The density of skid trails within the Baru catchment is

comparable to that in the Bukit Berembun C1 catchment

(Peninsular Malaysia) with its similar conventional selective

logging practices (Table 6). The Baru skid trail density is,

however, a factor of 2.8 larger than that of the reduced impact

logging (RIL) site at Bukit Berembun and a factor of 2.9 fold

smaller than that in the clearfelled Bukit Tarek catchment also

Table 7

DBM model parameters identified for the Baru stream (58010N, 117848.750E) nearinfiltration-excess according to road and skid trails areas, (b) undertaking a scenario

with (c) Malaysian clearfell forestry

Parameters & statistics DBM model

1a 1b

Structure [1 1 1] [1 1 4]

R2t

0.7322 0.715

YIC �9.492 �11.948

t 385 320

h 1415 1150

�0.4028 �0.8739

s() 0.0057 0.0006

P 0.0191 0.0884

s(P) 0.0002 0.0004

TC 5.499 37.10

s(TC) 0.085 0.199

ssP 0.0320 0.7009

RC* 0.0320 0.7006

Model 1a: DBM TF with BOSM non-linear model of net rainfall to 3.2% overlan

subsurface flow when overland flow is 3.2%;Model 2a: DBMTF with BOSM non-lin

TF with BOSM non-linear model of net rainfall to subsurface flow when overland

rainfall to overland flow for clearfell skid trails; Model 3b: DBM TF with BOSM n

(clearfell). See Table 3 for definition of parameters.

in Peninsular Malaysia (Abdul Rahim, 1990; Noguchi et al.,

2004; Table 6). Given this finding, two possible forestry

management ‘scenarios’ might be:

(1) W

the

with

2a

[1 1

0.73

�9.

385

141

�0.

0.00

0.01

0.00

5.49

0.08

0.02

0.02

d fl

ear

flow

on-li

hat is the impact of a skid trail density comparable to

Malaysian RIL forestry on the Baru stream hydrograph

(following 7-years of recovery)?

(2) W

hat is the impact of a skid trail density comparable to

Malaysian clearfell forestry on the Baru stream hydrograph

(following 7-years of recovery)?

If the statistical distribution of infiltration capacity of the

skid trails under RIL or clearfell operations is assumed to be

similar to that under the conventional selective logging

undertaken in the Baru during 1988/89, then the proportion

of infiltration-excess overland flow within the Baru can be

changed by simply altering the fscenario term, and hence b-term

(see Eq. (14)) in the gain elements for the overland and

subsurface flow partition. In both scenarios, fscenario was

changed to reflect the altered density of skid trails. If the

number of skid trails seen within a Malaysian RIL system had

been present within the Baru, but with the same soil

characteristics (‘scenario 1’), then the fractional area covered

by skid trails would reduce to 0.0143 (i.e. 0.04/2.8), so that the

proportion of net rainfall as overland flow (fscenario) over all

types of slope in the catchment would reduce to:

ð0:014 fractional area� 20%Þþ ð0:986 fractional area� 2:5%Þ¼ 2:75% of net rainfall (18)

In some contrast, if the number of skid trails seen within a

Malaysian clearfell system had been present within the Baru,

but with the same soil characteristics (‘scenario 2’), then the

Danum Valley Field Station (Malaysian Borneo), after (a) weighting the

the density of skid trails seen with Malaysian reduced impact logging, and

2b 3a 3b

1] [1 1 4] [1 1 1] [1 1 4]

2 0.726 0.732 0.678

492 �12.054 �9.492 �11.637

320 385 320

5 1169 1415 1132

4028 �0.8709 �0.4028 �0.8828

57 0.0006 0.0057 0.0007

64 0.0910 0.0271 0.0806

01 0.0004 0.0002 0.0004

9 36.180 5.499 40.10

5 0.188 0.085 0.24

75 0.7051 0.0453 0.6871

75 0.7051 0.0453 0.6873

ow; Model 1b: DBM TF with BOSM non-linear model of net rainfall to

model of net rainfall to overland flow given RIL skid trails; Model 2b: DBM

is 2.75% (RIL); Model 3a: DBM TF with BOSM non-linear model of net

near model of net rainfall to subsurface flow when overland flow is 4.53%

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–8074

Fig. 10. Simulated streamflow for the Baru catchment for the example storm

with a return period of 2 months (broken line) together with that predicted using

the density of skid trails from the reduced impact logging (RIL) system of Bukit

Berembun (solid line).

Fig. 11. Simulated streamflow for the Baru catchment for the example storm

with a return period of 2 months (broken line) together with that predicted using

the density of skid trails from the clearfell system of Bukit Tarek (solid line).

area covered by skid trails would increase to 0.116 (i.e.

0.04 � 2.9), so that the proportion of net rainfall as overland

flow (fscenario) over all types of slope in the catchment would

increase to:

ð0:116 fractional area � 20%Þþ ð0:884 fractional area� 2:5%Þ¼ 4:53% of net rainfall (19)

The resultant b-values for the RIL and clearfell scenarios

would be 0.859 (i.e. 2.75/3.2) and 1.416 (i.e. 4.53/3.2),

respectively. For these examples, only the effect of skid trail

density on the amount of net rainfall following a purely surface

pathway or purely subsurface pathway was simulated directly.

Clearly, a greater coverage of skid trails is likely to be

associated with greater canopy damage, so the rates of both

wet-canopy evaporation and transpiration would be different

also (Bidin, 2001; Chappell et al., 2001; Bidin et al., 2003). In

the two scenarios presented here, wet-canopy evaporation and

transpiration remain unchanged after altering the surface–

subsurface partition. Following future improvements to the

BARUMODEL evapo-transpiration components, we then aim

to examine the direct and indirect effects of forestry operations

on the wet-canopy evaporation and transpiration.

5.3. Results of the skid trail scenario modelling

The predicted streamflow results for ‘scenario 1: skid trail

density in a RIL system‘ are illustrated in Fig. 10, while those of

‘scenario 2: skid trail density in a clearfell system’ are illustrated

in Fig. 11. The scenario modelling was undertaken for the 12-

month period comprising of 105,408 values in each time-series,

but the effects are illustrated by plotting the hydrograph of a

storm with a return period of 2 months, i.e. ca. 50 mm/day (see

Douglas et al., 1999; Bidin, 2001). Figs. 10a and 11a show the

impact on the infiltration-excess overland flow component

during the example storm with a 2-month return period (28

September 1995) while Figs. 10b and 11b show the overall

impact on the streamflow during the same example storm. With

the Malaysian RIL scenario where the overall overland flow

proportion is reduced from 3.2 to 2.75% of net rainfall, the

overland flow in the ca. 50 mm event reduced by 16.38%. In

contrast, under theMalaysian clearfell scenariowhere the overall

overland flow proportion is increased from 3.2 to 4.53% of net

rainfall, the overland flow for the same event increased by

29.34%. Critically, the streamflow peak of the same event

actually increased marginally by 0.65% during the RIL

simulation, and reduced again marginally by 1.66% during

the clearfell simulation. These results are representative of the

changes seenwithin the other storms that comprise the 12-month

record. The perhaps unexpected result of a reducing stream

peakflow, although the small change is well within the

uncertainty of both the time-series data and modelling, arises

because the contribution of the overland flow is on the rising

stage, not the peak of the streamhydrograph (see Fig. 11b). Thus,

slightlymorewatermoves along the overlandflowpathway at the

start of the storm, leaving slightly less for the subsurface pathway

which makes up the majority of the streamflow peak. The

opposite result is seen with the RIL scenario.

6. Discussion

The aim of this study was to identify water pathways within

a humid tropical catchment using Data Based Mechanistic

(DBM) modelling, and then to apply the resultant models to

predict the effects of skid trail density within differing

rainforest management systems on the streamflow hydrograph.

A key underlying objective was to demonstrate the value of

further analysis of existing high quality time-series hydro-

logical data from tropical rainforest catchments. A linkage of

the component DBM models into a single structure for the

assessment of conceptual or data limitations and for subse-

quent scenario simulation we call the BARUMODEL. The

model should be considered to have a form which requires

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–80 75

development, but has already allowed us to newly consider

differences between water pathways, where new process

monitoring is required, and how we might devise new

experiments to evaluate forestry impacts within our equatorial

rainforest catchment. All of these issues have been aided by the

fact that the DBM-defined components of BARUMODEL are

clearly visible and can be readily modified as new time-series

data sets become available. This is a clear advantage of a model

based on observed time-series components over many physics-

based catchment models.

6.1. Hydrological interpretation of DBM model structures

and parameters

Even after more than 30 years of research into water

pathways within the wet tropics, there remains much debate

about: (a) the magnitude of infiltration-excess overland flow;

(b) the dominant pathway of water in the subsurface; and (c) the

best way of examining the controls on the runoff pathways (see

e.g. Bonell, 2004; Chappell et al., 1998, 2004b; McDonnell,

2003; Chappell, 2005; Chappell and Sherlock, 2005). Thinking

about tropical water pathways inferred from DBM modelling

has not been widely undertaken and, therefore, more DBM

modelling has the potential to add to the debate about the

tropical flow processes and their assessment. DBMmodels can:

(a) show differences in the non-linearity of water pathways; (b)

show whether distinctly different water pathways express

themselves in the streamflow records; (c) show differences in

the flashiness or residence times of catchments or individual

water pathways; (d) show relationships between variables not

previously considered locally; and (e) provide a robust basis for

forecasting future streamflow values.

Within this study, DBM modelling has shown that the

streamflow generation model is very poor when only linear

dynamics are considered, with a 28 percentage point increase in

variance explained where the BOSM non-linear filter is added

(Table 3). The strong non-linearity in the response relates to the

dominant subsurface pathway, specifically to the dynamic

effects of unsaturated zone storage (Young and Beven, 1994) or

the vertical distribution of permeability in the phreatic zone

(Kirkby, 1975; Chappell et al., 2004c). This contrasts with the

infiltration-excess overland flow response, which has a much

smaller 13 percentage point increase in variance explained

where the BOSM non-linear filter is added (Table 5). Thus, the

non-linearity in the infiltration-excess overland flow response

due to variations in ground-surface wetness (Horton, 1933),

exerts a much smaller influence on infiltration than do the water

content dynamics in the whole soil-rock profile on subsurface

drainage and exfiltration.

Some conceptual models, even those which are said to be

data-based (e.g. Croke et al., 2004), can assume the effects of

two distinct water pathways express themselves in the

streamflow response, and so use a model structure with two

pathways. The DBM modelling with its ability to use SRIV to

search for higher order models, indicates that no more than one

very dominant pathway has expression in the Baru stream

hydrograph. This does not mean that two or indeed many more

surface and subsurface pathways are absent with the catchment,

just that: (i) one pathway totally dominates the response; (ii) the

combined effect of several pathways has the appearance of a

first order response; or (iii) the DRCs, notably d and TC, of thecomponent pathways are not sufficiently different. Many

catchments where DBM modelling identifies a second order

model have a marked seasonality in a deep aquifer response

combined with a very flashy storm-related response in the

shallow subsurface. If the subsurface system does not express

such a higher order response in the streamflow records, but we a

priori assume a second order model, then uncertainty in our

DRCs or parameters increases, sometimes dramatically

(Young, 1992).

We expect that the infiltration-excess overland flow pathway

responds more quickly and for a shorter time than the

subsurface response which dominates the stream hydrograph.

This DBM study quantifies the difference in these dynamic

response characteristics using models derived from observed

time-series, possibly for the first time in a humid tropical forest,

and certainly for DBM modelling. The infiltration-excess

overland flow pathway has a shorter pure time delay before

initial response of 1 time-step (ca. 5 min: Table 5) compared to

the streamflow response which is 3 time-steps (ca. 15 min:

Table 3). Similarly this path recesses faster with a TC of 5.5 min

(Table 5) versus the very different 37 min of the overall

streamflow response (Table 3). The limited number of DBM

studies of runoff processes presented in Table 4 highlights the

need to undertake further DBM modelling work in the wet

tropical forests having a wide range of climatic and terrain

conditions. The more time-series that are modelled, the greater

the degree of interpretation of the DRCs that can be made. The

derivation of a subsurface flow time-series from a combination

of the streamflow and infiltration-excess overland flow time-

series is acceptable as a first approximation, given that the

streamflow is 96% subsurface flow. The need for this

approximation does, however, highlight the need to obtain

independent time-series of subsurface flow for subsequent

DBM modelling, perhaps from continuous tracer releases,

throughflow troughs or borehole data. We are, however, aware

of the current limitation of these three techniques in deriving

subsurface flow time-series (Knapp, 1970; Fawcett et al., 1997;

Chappell and Sherlock, 2005). While we know that the

infiltration-excess overland flow pathway responds faster than

the subsurface path with its longer route and greater storage,

having a simulation model derived directly from tropical

observations should be helpful for future research on tropical

runoff processes.

The DBM model of rising air temperature to transpiration

shows that there is a strong, dynamic relationship between these

two variables at least within the selected rainforest trees. Air

temperature is a commonly measuredmeteorological variable in

rainforest sites, so the model shows the potential for future

studies that might generate transpiration models from even short

runs of transpiration data. The transpiration is probably not

responding to air temperature, but to those meteorological

variables (e.g. net radiation) that also affect air temperature. We,

therefore, intend to extend our DBM modelling of transpiration

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–8076

by looking at its relationship with a range of hydro-

meteorological variables, notably wind speed, saturation deficit

of the atmosphere, soil capillary potential and net radiation (see

Roberts, 2000). We also intend to extend the DBMmodelling of

transpiration to sapflow data-series collected from other tree

species in the wet tropics (see e.g. Eschenbach et al., 1998) and

other transpiration measurement methods, e.g. eddy covariance

(see e.g. Kumagai et al., 2004). A Global Atmospheric Watch

(GAW) tower was recently constructed on a hill 2 km to the east

of the Baru catchment.We plan to undertake suchmeasurements

on this 100 mGAW tower. Critically, this tower has the potential

to allow us to measure catchment-lumped estimates of

transpiration, rather than trying to capture spatial variability

with many tree-scale measurements. Further DBMmodelling of

transpiration would then allow greater hydro-meteorological

interpretation of the DRCs which comprise the DBM transpira-

tion models. Equally, our planned instrumentation of the Baru

canopy with wetness sensors (i.e. 237 Wetness Sensing Grids

from Campbell Scientific Ltd., Shepshed, UK), may help us

understand transpiration suppression during storms, but also

allow us to develop a dynamic DBM model of wet canopy

evaporation. Wet canopy evaporation was poorly parameterised

as a long-term average value in this study (i.e. 1 � a, see Eq. (8)

and Fig. 3). Improved parameterisation of the evapo-transpira-

tion components should allow more complete scenarios of the

tropical forestry impacts including canopy and terrain changes to

be assessed.

6.2. BARUMODEL application to runoff impacts of skid

trail density

The BARUMODEL simulations indicated that while the

overland flow pathway reduced with Malaysian RIL and

increased with Malaysian clearfell systems during all storms

studied, this did not increase the stream hydrograph peaks

because most overland flow contributed before the peaks. The

fact that our scenario modelling indicates that stream peak flow

does not change significantly with a greater or lesser proportion

of overland flow also results from its relatively insignificant

contribution to the overall streamflow at our equatorial rainforest

site. This result, which goes against the popular perception of the

impacts of tropical forest skid trails on stream flashiness (see

Chappell, 2005) is, however, consistent with our modelling of

rainfall-runoff data for the Bukit Berembun catchments in

PeninsularMalaysia, just before and just after forestry operations

(Chappell et al., 2004b). In the case of the BARUMODEL

simulations presented here, it must be remembered that the time-

series data collected and analysed were from a period 7-years

post the initial phase of selective logging. As a result, the

overland flow behaviour monitored may be volumetrically less

significant than would have been the case during the logging

operations or immediately after (Douglas et al., 1995). It should

be remembered, however, that the clearfell scenario assumed that

overland flow on skid trails and roads amounted to 20% of net

rainfall, which is still are large proportion even for the logging

period (see Sinun et al., 1992). Further, wewould expect that the

less vegetated nature of skid trails during the logging period

would give greater overland flow volumes, and thus a quicker

response. Hence it is not expected that the peak in the overland

flow becomes more synchronous with the streamflow peak

thereby giving a significant impact, if anything, they would

become further apart.

With the two forestry scenarios illustrated here, we can see

that the streamflow hydrograph is dominated by waterflows in

non-road/trail parts of the catchment (see Eq. (17). We would

suggest that new quantitative data on surface water-flows (see

Chappell et al., 2004a), subsurface flows and properties such as

the soil infiltration capacity and permeability distribution (see

Chappell et al., 1998) from non-road or non-trail parts of the

Baru catchment are required urgently in this and other

rainforest catchments to better understand differences between

different tropical forestry systems.

The first episode of selective felling of tropical rainforests

has physical impacts resulting from changes to: (a) the forest

canopy; and (b) the terrain (Chappell et al., 2004b). Thus far,

we have only addressed the impacts of skid trails on the terrain

and not the canopy. Clearly, if the wet-canopy evaporation and

transpiration components in the BARUMODEL can be better

modelled, then changes to these pathways could be assessed.

While the scenario modelling indicates that the streamflow

peak is unaffected by different densities of skid trails typical of

different Malaysian logging systems, altering the proportion of

overland flow on disturbed surfaces, even by only a few percent

may have a very significant impact on the sediment delivery

(Douglas et al., 1992, 1999). Indeed, reviews by Bruijnzeel

(1990, 2001) and Chappell et al. (2004b) show that the greatest

physical impact of tropical forestry is on sediment yield. We,

therefore, need and intend to extend the BARUMODEL

approach to time-series sediment flux data from the nested

contributory areas in the Baru and thereby extend the sediment

analysis of Chappell et al. (1999, 2004a). Again, one of the

objectives would be to develop a model where the effect of

changing the relative contributions from different monitored

sources (e.g. landslides, culvert collapses, surface wash areas

on roads: see Chappell et al., 2004a) to the whole catchment

sediment yield can be assessed. This would allow a potentially

more significant forestry impact to be quantified and thereby

used by other to improve sustainable forestry guidelines for

equatorial rainforests.

7. Conclusions

The exercise of identifying DBM model structures and

DRCs from multiple water pathways in the same equatorial

forest catchment, and their combination to make a preliminary

assessment of varying skid trail density effects on runoff

response, yields six conclusions:

1. T

he relationship between equatorial rainfall and overland

flow or temperature and transpiration, sampled every 10 s

and integrated to a 5 min time-series, can be modelled using

a Single-Input-Single-Output (SISO) DBM approach. The

rainfall-HOF and temperature–transpiration models may

be the first such DBM models for equatorial rainforests.

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–80 77

This success allows the characteristics (DRCs) of these water

pathways to be compared with those of the lumped rainfall–

streamflow DBM model, and the potential for building a

multi-component catchment model where some components

have small impact on streamflow prior to any disturbance.

2. P

rocess hydrologists and forest hydrologists expect the

infiltration-excess overland flow pathway to be related to

rainfall in a more linear manner, respond more quickly and

for a shorter time than the subsurface response which

dominates the stream hydrograph. This study for the first

time quantifies these three characteristics (i.e. t, d, TC) for

overland flow in a wet equatorial rainforest, allowing

objective comparison with the lumped streamflow response

or a derived subsurface flow response (i.e. Q(t)-QQF(t)).

Thus, for the representative slope section generating

infiltration-excess overland flow, the three characteristics

are: 385, 1 and 5.5 for t, d and TC, respectively, while for the

response of the 0.44 km2 Baru catchment stream they are:

300, 3 and 37. These are considered to be relatively reliable

estimates of these hydrological DRCs. This reliability

uniquely comes from calibration of model structure and

parameters using the SRIValgorithm and because only those

component DBM models that have pathways having

expression in an output time-series, are considered

statistically acceptable.

3. A

new, but explicitly preliminary, DBM model of

transpiration shows that a Transfer Function with only

rising air temperature as the input explains 79% of the

variance in transpiration from individual rainforest trees.

Thus far, only 15-year-old camphor (D. lanceolata) trees

have been monitored and modelled. The preliminary

estimate of catchment-wide transpiration was made by

using an integrator which normalises the data for a

representative tree response using the difference in the

annual average net rainfall and streamflow.

4. M

odelling the streamflow generation pathways, by separate

DBM modelling of the component pathways rather than

DBM identification of only the dominant pathways

expressed in the streamflow records, gives a slightly less

accurate streamflow model, but one which allows volume-

trically insignificant component pathways to be studied and

altered. Our field data in the Baru catchment showed that the

infiltration-excess overland flow pathway contributed only a

very minor component of the streamflow hydrograph. The

representative overland flow site carried only 2.4% of the net

rainfall. Preliminary analysis of water balances for all

overland flow sites combined with an analysis of haulage

road and skid trail density in the Baru suggested that a figure

of 3.2% of net rainfall would be more representative. Thus,

the catchment model made by combining the DBMmodel of

overland flow with the DBM model of subsurface flow,

normalised the partition between long-term overland flow

and subsurface flow to 3.2 and 71.6%, respectively. Gain

elements using b and 1 � b, were then added to allow this

partition to be altered to simulate the effects of a different

proportion of slid trail surface with its representative value of

overland flow. These gain elements, combined with the

linked DBM overland flow and DBM subsurface flow

models and the DBM transpiration model, linked with an

integrator, we call the BARUMODEL.

5. R

eviewing published figures of skid trail density within

Malaysian rainforests affected by forestry, suggests the

Reduced Impact (RIL) forms of selective logging have a

factor of 2.8 fewer skid trails, while clearfell forms of

logging have a factor of 2.9 more skid trails. Weighting the

overland flow according to these different areas, but

assuming the same dynamics for a unit area of skid trail,

this gives 2.75 and 4.53% of net rainfall moving as overland

flow, respectively. Within BARUMODEL, the subsurface

flow is changed proportionately (i.e. 1 � a) to reflect the

changed overland flow, thus the simulated annual streamflow

total remains unchanged. The time distribution of streamflow

does, however, change. The most obvious change is to the

rising stage of hydrographs during larger rainstorms, e.g.

those with a return period of 2 months or ca. 50 mm/day.

Selecting one such storm for illustration, the simulation

showed that the peak overland flow reduced by 16% for the

Malaysian RIL system, and increased by 29% for the

Malaysian clearfell system. Critically, the simulated stream-

flow peak marginally increased with RIL and marginally

decreased with clearfelling, though both changes are within

the data and modelling uncertainty. This arose because the

overland flow contributes primarily to the rising stage, rather

than peak of each stream hydrograph. The insignificant

impact of overland flow on skid trails to the streamflow

hydrograph is consistent with our review and modelling

work elsewhere in Malaysia and the humid tropics and

supports the idea that the popular perception of the impacts

of tropical forestry on hydrology can be exaggerated in some

cases. This does not mean that even very small changes to the

overland flow pathway are not important. Erosion and

sediment delivery are very sensitive to even small changes in

overland flow within disturbed rainforest areas (Douglas

et al., 1995, 1999). This observation suggests that we should

extend our DBM-based scenario modelling to sediment

delivery in the Baru catchment where we have time-series

data representative of the dynamics of different sediment

sources.

6. A

ssessing and combining the different water pathway

components within the BARUMODEL has highlighted

several significant limitations of our existing data and hence

modelling capabilities. We need high frequency (e.g. a 5 min

time-base) sampling of wet canopy evaporation bymeasuring

canopy wetness in addition to the gross and net rainfall.

Transpiration dynamics more representative of catchment-

wide behaviour would come from: (a) sapflow measurements

on many different tree genera; (b) use of other meteorological

input variables; and (c) using eddy covariance measurements

from the 100 m tower by the Baru catchment in particular.

Many more measurements of infiltration-excess overland

flow, particularly in the more extensive non-road or non-trail

areas when combined with DBM modelling of all sites (see

Fig. 1), would improve the reliability of the catchment-

average DBMmodel of this surface pathway. We continue to

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–8078

search for time-series of subsurface flow that would be a better

representation of the below ground pathway rather than the

Q(t) � QOF(t) time-series used here. Developing a dynamic

partition between overland flow and subsurface flow variables

rather than the static (long-term) partition used here would

also improve the statistical and physical robustness of the

BARUMODEL. Similarly the effect of rainfall on transpira-

tion, namely transpiration suppression during rainstorms, will

be simulated as canopywetness and new sapflow data become

available, thus improving the BARUMODEL linkages. All of

these issues highlight the need for muchmore time-series data

collection within wet equatorial rainforests managed for

timber production. TheDBMmodelling should be considered

a method to identify dynamic linkages between hydrological

variables and data limitations and thus areas for new research,

rather than a conclusion to field research.

Acknowledgements

In particular, the authors wish to acknowledge the

considerable efforts of Mohamed Jamal Hanapi, Johnny

Larenus and Paun Kukuon who have had the primary

responsibility for the field data collection in the Baru catchment

over the last 10 years. The considerable support of Yayasan

Sabah (Forestry Division and R&D Division) and the Danum

Valley Management Committee has been invaluable to all

aspects of our work. Others making a significant contribution to

the success of this particular study are the Royal Society of

London (South East Asia Rainforest Research Programme) and

the Danum Survey Team (Ismail Abdul Samat, Shaidih Abdul

Samat and Nick Chappell). Comments from Prof Keith Beven

(Lancaster University), issue co-editor (Prof Tani, Kyoto

University) and manuscript reviewers are gratefully acknowl-

edged. Discussions on various aspects of this work with

participants at the IUFRO Forest Hydrology workshop in Kota

Kinabalu, Sabah, 10–15th July 2004 was most valuable and

thus much appreciated.

References

Abdul Rahim, N., 1990. The effects of selective logging methods on hydro-

logical parameters in Peninsular Malaysia. PhD Thesis, University of

Wales, Bangor, unpublished.

Baharuddin, K., 1995. Effects of properties and soil erosion in a hill dipterocarp

forest of Peninsular Malaysia. Research Pamphlet No. 119. Forestry

Research Institute of Malaysia, Kuala Lumpur.

Bathurst, J.C., Ewen, J., Parkin, G., O’Connell, P.E., Cooper, J.D., 2004.

Validation of catchment models for predicting land-use and climate change

impacts. 3. Blind validation for internal and outlet responses. J. Hydrol. 287,

74–94.

Beven, K.J., 2000. Uniqueness of place and process representations in hydro-

logical modelling. Hydrol. Earth Syst. Sci. 4, 203–214.

Beven, K., 2001a. On fire and rain (or predicting the effects of change). Hydrol.

Process. 15, 1397–1399.

Beven, K., 2001b. How far can we go in distributed hydrological modelling?

Hydrol. Earth Syst. Sci. 5, 1–12.

Beven, K.J., 2002. Towards an alternative blueprint for a physically-based

digitally simulated hydrologic response modelling system. Hydrol. Process.

16, 189–206.

Beven, K., Freer, J., 2001. Equifinality, data assimilation, and uncertainty

estimation in mechanistic modelling of complex environmental systems

using the GLUE methodology. J. Hydrol. 249, 11–29.

Bidin, K., 2001. Spatio-temporal variability in rainfall and wet-canopy eva-

poration within a small catchment recovering from selective tropical

forestry. PhD Thesis. University of Lancaster, Lancaster (UK), unpublished.

Bidin, K., Chappell, N.A., 2003. First evidence of a structured and dynamic

spatial pattern of rainfall within a small humid tropical catchment. Hydrol.

Earth Syst. Sci. 7, 245–253.

Bidin, K., Chappell, N.A., 2006. Characteristics of rain-events at an inland

locality in Northeastern Borneo. Hydrol. Process., in press.

Bidin, K., Chappell, N.A., Sinun, W., Tangki, H., 2003. Net-rainfall and wet-

canopy evaporation in a small selectively-logged rainforest catchment,

Sabah, Malaysia. In: Proceedings of First International Conference on

Hydrology and Water Resources in Asia Pacific Region, vol. 2, pp. 659–

666.

Bonell, M., 2004. Runoff generation in tropical forests. In: Bonell, M.,

Bruijnzeel, L.A. (Eds.), Forests, Water and People in the Humid Tropics.

Cambridge University Press, Cambridge, pp. 314–406.

Box, G.E.P., Jenkins, G.M., 1970. Time Series Analysis: Forecasting and

Control. Holden-Day, San Francisco.

Bruijnzeel, L.A., 1990. Hydrology of Moist Tropical Forest and Effects of

Conversion: A State of the Art Review. UNESCO, Paris.

Bruijnzeel, L.A., 1992. Managing tropical forest watersheds for production:

where contradictory theory and practice co-exist. In: Miller, F.R., Adam,

K.L. (Eds.), Wise Management of Tropical Forests, Proceedings of the

Oxford Conference on Tropical Forests, 1992. Oxford Forestry Institute

(University of Oxford), Oxford, pp. 37–75.

Bruijnzeel, L.A., 2001. Forest hydrology. In: Evans, J. (Ed.), The Forests

Handbook. Blackwell, Oxford, pp. 301–343.

Bruijnzeel, L.A., 2004. Hydrological functions of tropical forests: not seeing the

soil for the trees? Agric. Ecosyst. Environ. 104, 185–228.

Camarasa, A.M., Tilford, K.A., 2002. Rainfall-runoff modelling of ephemeral

streams in the Valencia region (eastern Spain). Hydrol. Process. 16, 3329–

3344.

Chappell, N.A., 2005.Water pathways in humid forests: myths vs. observations.

Suiri Kagaku (Water Sci.) 48, 32–46.

Chappell, N.A., Fowell, M., 2003. Evaluation of GCM simulation of humid

tropical hydrology. UGAMP Newslett. 27, 8–9.

Chappell, N.A., Sherlock, M.D., 2005. Contrasting flow pathways within

tropical forest slopes of Ultisol soil. Earth Surf. Process. Landforms 30,

735–753.

Chappell, N.A., Ternan, J.L., 1992. Flow-path dimensionality and hydrologic

modelling. Hydrol. Process. 6, 327–345.

Chappell, N.A., Franks, S.W., Larenus, J., 1998. Multi-scale permeability

estimation for a tropical catchment. Hydrol. Process. 12, 1507–1523.

Chappell, N.A., McKenna, P., Bidin, K., Douglas, I., Walsh, R.P.D., 1999.

Parsimonious modelling of water and suspended-sediment flux from nested-

catchments affected by selective tropical forestry. Philos. Trans. Royal Soc.

Lond. Ser. B. 354, 1831–1846.

Chappell, N.A., Bidin, K., Tych, W., 2001. Modelling rainfall and canopy

controls on net-precipitation beneath selectively-logged tropical forest.

Plant Ecol. 153, 215–229.

Chappell, N.A., Douglas, I., Hanapi, J.M., Tych, W., 2004a. Source of sus-

pended-sediment within a tropical catchment recovering from selective

logging. Hydrol. Process. 18, 685–701.

Chappell, N.A., Nik, A.R., Yusop, Z., Tych, W., Kasran, B., 2004b. Spatially-

significant effects of selective tropical forestry on water, nutrient and

sediment flows: a modelling-supported review. In: Bonell, M., Bruijn-

zeel, L.A. (Eds.), Forests, Water and People in the Humid Tropics. Cam-

bridge University Press, Cambridge, pp. 513–532.

Chappell, N.A., Bidin, K., Sherlock, M.D., Lancaster, J.W., 2004c. Parsimo-

nious spatial representation of tropical soils within dynamic, rainfall-runoff

models. In: Bonell, M., Bruijnzeel, L.A. (Eds.), Forests, Water and People

in the Humid Tropics. Cambridge University Press, Cambridge, pp. 756–

769.

Conway, S., 1982. Logging Practices. Miller Freeman Publications, San

Francisco.

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–80 79

Croke, B.F.W., Merritt, W.S., Jakeman, A.J., 2004. A dynamic model for

predicting hydrologic response to land cover changes in gauged and

ungauged catchments. J. Hydrol. 291, 115–131.

Douglas, I., Spencer, T., Greer, T., Bidin, K., Sinun,W., Wong,W.M., 1992. The

impact of selective commercial logging on stream hydrology, chemistry and

sediment loads in the Ulu Segama rain forest, Sabah. Philos. Trans. Roy.

Soc. Lond. Ser. B 335, 397–406.

Douglas, I., Greer, T., Sinun, W., Anderton, S., Bidin, K., Spilsbury, M., Jadda,

S., Azman, S., 1995. Geomorphology and rainforest logging practices. In:

McGregor, D.F.M., Thompson, D.A. (Eds.), Geomorphology and Land

Management in a Changing Environment. Wiley, Chichester.

Douglas, I., Bidin, K., Balamurugam, G., Chappell, N.A., Walsh, R.P.D., Greer,

T., Sinun, W., 1999. The role of extreme events in the impacts of selective

tropical forestry on erosion during harvesting and recovery phases at Danum

Valley, Sabah. Phil. Trans. Roy. Soc. Lond. B. 354, 1749–1761.

Dubreuil, P.L., 1985. Review of field observations of runoff generation in the

tropics. J. Hydrol. 80, 237–264.

Elliot, W.J., Tysdal, L.M., 1999. Understanding and reducing erosion from

insloping roads. J. Forest. 97, 30–34.

Eschenbach, C., Glauner, R., Kleine, M., Kappen, L., 1998. Photosynthesis

rates of selected tree species in lowland dipterocarp rainforest of Sabah,

Malaysia. Trees-Struct. Funct. 12, 356–365.

Fawcett, C.P., Young, P.C., Feyen, H., 1997. Validation of Data-Based Mechan-

istic non-linear rainfall-flow model. In: Proceedings of the Twelfth Inter-

national Conference on Systems Engineering, ICSE’97, Coventry

University, UK, 9-11 September 1997, Coventry University, Coventry,

pp. 252–257.

Forestry Department Peninsular Malaysia, 1999. Specification of Forest Roads

for Peninsular Malaysia. Forestry Department Peninsular Malaysia, Kuala

Lumpur.

Gregory, K.J., Walling, D.E., 1973. Drainage Basin Form and Process. Arnold,

London.

Horton, R.E., 1933. The role of infiltration in the hydrologic cycle. Trans. Am.

Geophys. Union 14, 446–460.

Hundecha, Y., Bardossy, A., 2004.Modeling of the effect of land use changes on

the runoff generation of a river basin through parameter regionalization of a

watershed model. J. Hydrol. 292, 281–295.

Jones, J.A., Grant, G.E., 1996. Peak flow responses to clear-cutting and roads in

small and large basins, western Cascades, Oregon.Water Resources Res. 32,

959–974.

Kirkby, M.J., 1975. Hydrograph modelling strategies. In: Peel, R., Chisholm,

M., Haggett, P. (Eds.), Process in Physical, Human, Geography.

Heinemann, pp. 69–90.

Knapp, B.J., 1970. Patterns of water movement on a steep upland hillside,

Plynlimon, central Wales. PhD Thesis, University of Reading, unpublished.

Kumagai, T., Katul, G.G., Saitoh, T.M., Sato, Y., Manfroi, O.J., Morooka, T.,

Ichie, T., Kuraji, K., Suzuki, M., Porporato, A., 2004. Water cycling in a

Bornean tropical rain forest under current and projected precipitation

scenarios. Water Resources Res. 40 (1) art. no. W01104.

Klemes, V., 2000. Common Sense and Other Heresies: Selected Papers on

Hydrology and Water Resources Engineering. Canadian Water Resources

Association, Cambridge.

Kuraji, K., 1996. Hydrological characteristics of moist tropical forests. Bull.

Tokyo Univ. Forests 95, 93–208.

Malmer, A., 1996. Hydrological effects and nutrient losses of forest plantation

establishment on tropical rainforest land in Sabah, Malaysia. J. Hydrol. 174,

129–148.

McDonnell, J.J., 2003. Where does water go when it rains? Moving beyond the

variable source area concept of rainfall-runoff response. Hydrol. Process.

17, 1869–1875.

Moreira, M.Z., Sternberg, L.D.L., Martinelli, L.A., Victoria, R.L., Barbosa,

E.M., Bonates, L.C.M., Nepstad, D.C., 1997. Contribution of transpiration

to forest ambient vapour based on isotopic measurements. Global Change

Biol. 3, 439–450.

Moura-Costa, P., Karolus, A., 1992. Timber extraction volumes (1970–1991) of

Ulu Segama Forest Reserve. Innoprise Corporation SdnBhd., Kota Kinabalu.

Moura-Costa, P.H., Yap, S.W., Ong, C.L., Ganing, A., Nussbaum, R., Mojiun,

T., 1996. Large scale enrichment planting with dipterocarps as an alternative

for carbon offset -methods and preliminary results. In: Appanah, S., Khoo,

K.C. (Eds.), Proceedings of the Fifth Round Table Conference on Dipter-

ocarps, ChiangMai, Thailand, November 1994, Forest Research Institute of

Malaysia (FRIM), Kepong, pp. 386–396.

Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual

models. Part I. A discussion of principles. J. Hydrol. 10, 282–290.

Noguchi, S., RahimNik, A., Tani,M., 2003. Rainfall characteristics of tropical

rainforest at Pasoh Forest Reserve, Negeri Sembilan, Peninsular Malaysia.

In: Okuda, T., Manokaran, N., Matsumoto, Y., Niiyama, K., Thomas,

S.C., Ashton, P.S. (Eds.), Pasoh: Ecology of a Lowland Rain Forest in

Southeast Asia. Springer, Tokyo, pp. 51–58.

Noguchi, S., Abdul Rahim, N., Negeshi, J.N., Sasaki, S., Abdul Salam, A.C.,

2004. Overland flow pathways at Hopea Odorata Plantation in Peninsular

Malaysia. In: Sidle, R.C., Tani, M., Nik, A.R., Tadese, T.A. (Eds.), Forests

and Water in Warm, Humid Asia. Disaster Prevention Research Institute,

Uiji, pp. 208–210.

Nussbaum, R., Anderson, J.M., Spencer, T., 1995. Factors limiting the growth of

indigenous tree saplings planted on degraded rainforest soils in Sabah,

Malaysia. For. Ecol. Manage. 74, 149–159.

Ott, B., Uhlenbrook, S., 2004. Quantifying the impact of land-use changes at the

event and seasonal time scale using a process-oriented catchment model.

Hydrol. Earth Syst. Sci. 8, 62–78.

Parkinson, S.D., Young, P.C., 1998. Uncertainty and sensitivity in global carbon

cycle modelling. Clim. Res. 9, 157–174.

Pinard, M.A., Putz, F.E., Tay, J., Sullivan, T.E., 1995. Creating Timber Harvest

Guidelines for a Reduced-Impact Logging project inMalaysia. J. Forest. 93,

411–445.

Pinard, M.A., Barker, M.G., Tay, J., 2000. Soil disturbance resulting from

bulldozer-yarding of logs and post-logging forest recovery on skid trails in

Sabah, Malaysia. For. Ecol. Manage. 130, 213–225.

Rhee, H., Fridley, J.L., Foltz, R.B., 2004. Modeling erosion from unpaved forest

roads at various levels of geometric detail using the WEPP model. Trans.

ASAE 47, 961–968.

Roberts, J., 2000. The influence of physical and physiological characteristics

dof vegetation on their hydrological response. Hydrol. Process. 14, 2885–

2901.

Sidle, R.C., Sasaki, S., Otsuki, M., Noguchi, S., Nik, A.R., 2004. Sediment

pathways in a tropical forest: effects of logging roads and skid trails. Hydrol.

Process. 18, 703–720.

Sherlock, M.D., Chappell, N.A., McDonnell, J.J., 2000. Effects of experimental

uncertainty on the calculation of hillslope flow paths. Hydrol. Process. 14,

2457–2471.

Sinun, W. 1991. Hillslope hydrology, hydrogeomorphology and hydrochem-

istry of an equantorial lowland rainforest, Danum Valley, Sabah, Malaysia.

PhD Thesis, University of Manchester, Manchester (UK), unpublished.

Sinun, W., Meng, W.W., Douglas, I., Spencer, T., 1992. Throughfall, stemflow,

overland flow and throughflow in the Ulu Segama rain forest, Sabah,

Malaysia. Philos. Trans. Roy. Soc. Lond. Ser. B. 335, 389–395.

Thang, H.C., 1987. Forest management-systems for tropical high forest, with

special reference to Peninsular Malaysia. For. Ecol. Manage. 21, 3–20.

Thang, H.C., Chappell, N.A., 2004. Minimising the hydrological impact of

forest harvesting in Malaysia’s rainforests. In: Bonell, M., Bruijnzeel, L.A.

(Eds.), Forests, Water and People in the Humid Tropics. Cambridge

University Press, Cambridge, pp. 852–865.

Thomas, R.B., Megahan, W.F., 1998. Peak flow responses to clear-cutting and

roads in small and large basins, western Cascades, Oregon. A second

opinion. Water Resources Res. 34, 3393–3403.

Tukey, J.W., 1961. Discussion emphasizing the connection between analysis of

variance and spectrum analysis. Technometrics 3, 191.

Van der Plas, M.C., Bruijnzeel, L.A., 1993. The impact of mechanized selective

logging of lowland rain forest on topsoil infiltrability in the upper Segama

area, Sabah, Malaysia. International Association of Hydrological Sciences.

Publication no. 216, pp. 203–211.

Vertessy, R.A., Elsenbeer, H., 1999. Distributed modeling of stormflow gen-

eration in an Amazonian rain forest catchment: effects of model parame-

terization. Water Resources Res. 35, 2173–2187.

Viney, N.R., Sivapalan, M., 2001. Modelling catchment processes in the Swan-

Avon River Basin. Hydrol. Process. 15, 2671–2685.

N.A. Chappell et al. / Forest Ecology and Management 224 (2006) 58–8080

Whitehead, P.G., Wilson, E.J., Butterfield, D., Seed, K., 1998. A semi-dis-

tributed Integrated Nitrogen model for multiple source assessment in

Catchments (INCA): Part II-application to large river basins in south Wales

and eastern England. Sci. Total Environ. 210/211, 559–583.

Young, P.C., 1984. Recursive Estimation and Time-Series Analysis. Springer–

Verlag, Berlin.

Young, P.C., 1985. The instrumental variable method: a practical approach to

identification and system parameter estimation. In: Barker, H.A., Young,

P.C. (Eds.), Identification and System Parameter Estimation. Pergamon,

Oxford, pp. 1–16.

Young, P.C., 1992. Parallel processes in hydrology and water quality. J. Inst.

Water Environ. Manage. 6, 598–612.

Young, P.C., 1999. Data-Based Mechanistic modelling, generalized sensitivity

and dominant mode analysis. Comput. Phys. Commun. 115, 1–17.

Young, P.C., 2001. Data-Based Mechanistic modelling and validation of rain-

fall-flow processes. In: Anderson, M.G., Bates, P.D. (Eds.), Model Valida-

tion: Perspectives in Hydrological Science.Wiley, Chichester, pp. 117–161.

Young, P., 2003. Top-down and data-based mechanistic modelling of rainfall-

flow dynamics at the catchment scale. Hydrol. Process. 17, 2195–2217.

Young, P.C., Beven, K., 1994. Data-based mechanistic modelling and the

rainfall-flow non-linearity. Environmetrics 5, 335–363.

Young, P.C., Chotai, A., Beven, K.J., 2004. Data-Based Mechanistic modelling

and the simplification of environmental systems. In: Wainright, J., Mulli-

gan, M. (Eds.), Environmental Modelling: Finding Simplicity in Complex-

ity. Wiley, Chichester, pp. 371–388.

Ziegler, A.D., Giambelluca, T.W., Sutherland, R.A., Vana, T.T., Nullet, M.A.,

2001. Horton overland flow contribution to runoff on unpaved mountain

roads. A case study in northern Thailand. Hydrol. Process. 15, 3203–3208.